{"generated_at":"2026-07-22T05:51:31.432789+00:00","min_benchmarks":7,"total_benchmarks_used":5206,"total_models":1776,"lower_is_better":["AIM-Bench (inventory manager)","Agent Arena - Tool Hallucination","AudioBench - AISHELL ASR ZH WER","AudioBench - Common Voice 15 EN WER","AudioBench - Earnings21 WER","AudioBench - Earnings22 WER","AudioBench - GigaSpeech WER","AudioBench - IMDA Part 1 ASR WER","AudioBench - IMDA Part 2 ASR WER","AudioBench - IMDA Part 3 ASR WER","AudioBench - IMDA Part 4 ASR WER","AudioBench - IMDA Part 5 ASR WER","AudioBench - IMDA Part 6 ASR WER","AudioBench - LibriSpeech Clean WER","AudioBench - LibriSpeech Other WER","AudioBench - SEAME Mandarin WER","AudioBench - SEAME Singapore English WER","AudioBench - TED-LIUM 3 Long Form WER","AudioBench - TED-LIUM 3 WER","AudioBench - YouTube ASR Batch 1 WER","AudioBench - YouTube ASR Batch 2 WER","ConStory-Bench","Confabulation Leaderboard (Lechmazur)","Deception Resistance (Lechmazur)","DystopiaBench","EffiBench - NET","EffiBench - NMU","FutureSearch BTF-2","FutureSearch BTF-2 Calibration","Generalization V1 (Lechmazur)","Guesswork","HalluHard","LLM Hallucination Detection - Non-RAG","LLM Hallucination Detection - RAG","MT-JailBench","Medical STT Benchmark","Open Universal Arabic ASR Leaderboard","Open Universal Arabic ASR Leaderboard - CER","PlatinumBench (MIT)","Position Bias (Lechmazur)","Sycophancy (Lechmazur)","Vector Eval - AgentHarm"],"benchmark_meta":{"100Q-Hard Net Score":{"url":"https://benchmarklist.com/benchmarks/100q_hard_net_score/","metric":"Net score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"A Matter of TASTE":{"url":"https://benchmarklist.com/benchmarks/a_matter_of_taste_improving_coverage_and_difficulty_of_agent_benchmarks/","metric":"Airline (Pass@1) τBV (self-reported)","theoretical_min":0.0,"theoretical_max":null},"A2RBench":{"url":"https://benchmarklist.com/benchmarks/a2rbench/","metric":"Total Acc (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AA AIME 2025":{"url":"https://artificialanalysis.ai/evaluations/aime-2025","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA APEX-Agents":{"url":"https://artificialanalysis.ai/evaluations/apex-agents-aa","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA CritPt":{"url":"https://artificialanalysis.ai/evaluations/critpt","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"sources":[{"name":"AA CritPt","url":"https://artificialanalysis.ai/evaluations/critpt","metric":"Accuracy (%)"},{"name":"CritPt (Fable/Mythos)","url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Score (%)"}]},"AA GDPval":{"url":"https://artificialanalysis.ai/evaluations/gdpval-aa","metric":"ELO","theoretical_min":null,"theoretical_max":null},"AA GPQA Diamond":{"url":"https://artificialanalysis.ai/evaluations/gpqa-diamond","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":25},"AA Global-MMLU-Lite":{"url":"https://artificialanalysis.ai/evaluations/global-mmlu-lite","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Global-MMLU-Lite - Arabic":{"url":"https://artificialanalysis.ai/evaluations/global-mmlu-lite","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Global-MMLU-Lite - Bengali":{"url":"https://artificialanalysis.ai/evaluations/global-mmlu-lite","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Global-MMLU-Lite - Burmese":{"url":"https://artificialanalysis.ai/evaluations/global-mmlu-lite","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Global-MMLU-Lite - Chinese":{"url":"https://artificialanalysis.ai/evaluations/global-mmlu-lite","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Global-MMLU-Lite - English":{"url":"https://artificialanalysis.ai/evaluations/global-mmlu-lite","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Global-MMLU-Lite - French":{"url":"https://artificialanalysis.ai/evaluations/global-mmlu-lite","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Global-MMLU-Lite - German":{"url":"https://artificialanalysis.ai/evaluations/global-mmlu-lite","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Global-MMLU-Lite - Hindi":{"url":"https://artificialanalysis.ai/evaluations/global-mmlu-lite","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Global-MMLU-Lite - Indonesian":{"url":"https://artificialanalysis.ai/evaluations/global-mmlu-lite","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Global-MMLU-Lite - Italian":{"url":"https://artificialanalysis.ai/evaluations/global-mmlu-lite","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Global-MMLU-Lite - Japanese":{"url":"https://artificialanalysis.ai/evaluations/global-mmlu-lite","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Global-MMLU-Lite - Korean":{"url":"https://artificialanalysis.ai/evaluations/global-mmlu-lite","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Global-MMLU-Lite - Portuguese":{"url":"https://artificialanalysis.ai/evaluations/global-mmlu-lite","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Global-MMLU-Lite - Spanish":{"url":"https://artificialanalysis.ai/evaluations/global-mmlu-lite","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Global-MMLU-Lite - Swahili":{"url":"https://artificialanalysis.ai/evaluations/global-mmlu-lite","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Global-MMLU-Lite - Yoruba":{"url":"https://artificialanalysis.ai/evaluations/global-mmlu-lite","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Humanity's Last Exam":{"url":"https://artificialanalysis.ai/evaluations/humanitys-last-exam","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA IFBench":{"url":"https://artificialanalysis.ai/evaluations/ifbench","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA LiveCodeBench":{"url":"https://artificialanalysis.ai/evaluations/livecodebench","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Long Context Reasoning":{"url":"https://artificialanalysis.ai/evaluations/artificial-analysis-long-context-reasoning","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA MATH-500":{"url":"https://artificialanalysis.ai/evaluations/math-500","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA MMLU-Pro":{"url":"https://artificialanalysis.ai/evaluations/mmlu-pro","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":10},"AA MMMU-Pro":{"url":"https://artificialanalysis.ai/evaluations/mmmu-pro","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Omniscience":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Score","theoretical_min":null,"theoretical_max":null},"AA Omniscience - Business":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Omniscience - Health":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Omniscience - Humanities & Social Sciences":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Omniscience - Law":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Omniscience - Science, Engineering & Mathematics":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Omniscience - Software Engineering (SWE)":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Omniscience - Software Engineering (SWE) - C":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Omniscience - Software Engineering (SWE) - Dart":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Omniscience - Software Engineering (SWE) - Go":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Omniscience - Software Engineering (SWE) - HTML":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Omniscience - Software Engineering (SWE) - Java":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Omniscience - Software Engineering (SWE) - JavaScript":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Omniscience - Software Engineering (SWE) - Julia":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Omniscience - Software Engineering (SWE) - Kotlin":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Omniscience - Software Engineering (SWE) - PHP":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Omniscience - Software Engineering (SWE) - Python":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Omniscience - Software Engineering (SWE) - R":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Omniscience - Software Engineering (SWE) - Rust":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Omniscience - Software Engineering (SWE) - Swift":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Omniscience - Software Engineering (SWE) - TypeScript":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA SciCode":{"url":"https://artificialanalysis.ai/evaluations/scicode","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA TAU-2 Bench":{"url":"https://artificialanalysis.ai/evaluations/tau2-bench","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA Terminal-Bench Hard":{"url":"https://artificialanalysis.ai/evaluations/terminalbench-hard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA-Briefcase":{"url":"https://artificialanalysis.ai/evaluations/aa-briefcase","metric":"Rubric Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA-LCR":{"url":"https://benchmarklist.com/benchmarks/aa_lcr/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AA-Omniscience Accuracy":{"url":"https://artificialanalysis.ai/evaluations/omniscience","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AA-Omniscience Net Score":{"url":"https://benchmarklist.com/benchmarks/aa_omniscience_net_score/","metric":"Net score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ABC-Bench":{"url":"https://benchmarklist.com/benchmarks/abc_bench/","metric":"Overall pass@1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ABRA":{"url":"https://benchmarklist.com/benchmarks/abra_agent_benchmark_for_radiology_applications/","metric":"Overall Avg (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ACEBench":{"url":"https://benchmarklist.com/benchmarks/acebench/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ADBench":{"url":"https://benchmarklist.com/benchmarks/ad_bench/","metric":"Pass@3 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AGC-Bench":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Mean JRT z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - Brainstorming":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"JRT z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - Figurative Language":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"JRT z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - Humor":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"JRT z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - Problem Solving":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"JRT z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - STEM":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"JRT z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - Story / Narrative":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"JRT z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - analobench":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - arastories":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - arn":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - brainteaser":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - c3_crosstalk":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - chinese_homophonic_puns":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - conceptual_design":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - cpers":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - creatset":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - crowd_vote":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - cue_word_story":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - data_narrative":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - fann_or_flop":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - fig_qa":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - future_ideas":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - futuregen":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - grapheval_ai_researcher":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - grapheval_iclr":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - grapheval_review_advisor":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - historical_analogy":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - humor_transfer":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - hypobench":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - hypogen":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - irfl":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - lcc_metaphor":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - liveideabench":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - meta4xnli":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - metaphor_generation":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - metaphoric_analogies":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - moh_x":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - mops":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - munch":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - newyorker_humor":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - ocw":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - ocw_connections":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - outline_to_story":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - permpst":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - poetmt":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - pollux_creativity":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - pron_vs_prompt":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - proparalogy":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - pun_eval":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - puntuguese":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - riddlesense":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - rpgbench":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - scar":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - schnovel":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - science_analogies":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - scimon":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - sdat":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - showerthoughts":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - simile_generation":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - slang_generation":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - speak_to_structure":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - ss_gen":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - story_generation_rocstories":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - story_quality":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - thenextchapter":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - tinyfabulist":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - tinystories":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - ttcw":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - twistlist":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGC-Bench - unfun_corpus":{"url":"https://huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","metric":"Dataset z-score","theoretical_min":null,"theoretical_max":null},"AGORA":{"url":"https://benchmarklist.com/benchmarks/agora_an_archive_grounded_benchmark_for_agentic_workplace_document_reasoning/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AI Chess Leaderboard (Continuation)":{"url":"https://dubesor.de/chess/chess-leaderboard","metric":"Elo","theoretical_min":null,"theoretical_max":null},"AI Chess Leaderboard (Reasoning)":{"url":"https://dubesor.de/chess/chess-leaderboard","metric":"Elo","theoretical_min":null,"theoretical_max":null},"AI Energy Score (Reasoning)":{"url":"https://huggingface.co/spaces/AIEnergyScore/Leaderboard","metric":"Energy Score (1-5)","theoretical_min":1.0,"theoretical_max":5.0},"AI Energy Score (Text Generation)":{"url":"https://huggingface.co/spaces/AIEnergyScore/Leaderboard","metric":"Energy Score (1-5)","theoretical_min":1.0,"theoretical_max":5.0},"AI for Education Pedagogy":{"url":"https://www.fab-ai.org/initiatives/ai-for-education/edtech-quality/benchmarks?benchmark=CDPK&cost=10&models=%5B%5D&pageView=table","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AI for Education Pedagogy - Maths":{"url":"https://www.fab-ai.org/initiatives/ai-for-education/edtech-quality/benchmarks?benchmark=CDPK&cost=10&models=%5B%5D&pageView=table","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AI for Education Pedagogy - Primary":{"url":"https://www.fab-ai.org/initiatives/ai-for-education/edtech-quality/benchmarks?benchmark=CDPK&cost=10&models=%5B%5D&pageView=table","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AI for Education Pedagogy - Science":{"url":"https://www.fab-ai.org/initiatives/ai-for-education/edtech-quality/benchmarks?benchmark=CDPK&cost=10&models=%5B%5D&pageView=table","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AI for Education Pedagogy - Secondary":{"url":"https://www.fab-ai.org/initiatives/ai-for-education/edtech-quality/benchmarks?benchmark=CDPK&cost=10&models=%5B%5D&pageView=table","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AI for Education Pedagogy - Social studies":{"url":"https://www.fab-ai.org/initiatives/ai-for-education/edtech-quality/benchmarks?benchmark=CDPK&cost=10&models=%5B%5D&pageView=table","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AI for Education Pedagogy - Technology":{"url":"https://www.fab-ai.org/initiatives/ai-for-education/edtech-quality/benchmarks?benchmark=CDPK&cost=10&models=%5B%5D&pageView=table","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AI for Education SEND":{"url":"https://www.fab-ai.org/initiatives/ai-for-education/edtech-quality/benchmarks?benchmark=CDPK&cost=10&models=%5B%5D&pageView=table","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AI for Education Visual Maths":{"url":"https://www.fab-ai.org/initiatives/ai-for-education/edtech-quality/benchmarks?benchmark=CDPK&cost=10&models=%5B%5D&pageView=table","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AI for Education Visual Maths - Algebra":{"url":"https://www.fab-ai.org/initiatives/ai-for-education/edtech-quality/benchmarks?benchmark=CDPK&cost=10&models=%5B%5D&pageView=table","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AI for Education Visual Maths - Geometry":{"url":"https://www.fab-ai.org/initiatives/ai-for-education/edtech-quality/benchmarks?benchmark=CDPK&cost=10&models=%5B%5D&pageView=table","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AI for Education Visual Maths - Measurement":{"url":"https://www.fab-ai.org/initiatives/ai-for-education/edtech-quality/benchmarks?benchmark=CDPK&cost=10&models=%5B%5D&pageView=table","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AI for Education Visual Maths - Number and Operations":{"url":"https://www.fab-ai.org/initiatives/ai-for-education/edtech-quality/benchmarks?benchmark=CDPK&cost=10&models=%5B%5D&pageView=table","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AI for Education Visual Maths - Statistics and Probability":{"url":"https://www.fab-ai.org/initiatives/ai-for-education/edtech-quality/benchmarks?benchmark=CDPK&cost=10&models=%5B%5D&pageView=table","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AI for Education Visual Reasoning":{"url":"https://www.fab-ai.org/initiatives/ai-for-education/edtech-quality/benchmarks?benchmark=CDPK&cost=10&models=%5B%5D&pageView=table","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AI for Education Visual Reasoning - match (figure)":{"url":"https://www.fab-ai.org/initiatives/ai-for-education/edtech-quality/benchmarks?benchmark=CDPK&cost=10&models=%5B%5D&pageView=table","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AI for Education Visual Reasoning - match (process)":{"url":"https://www.fab-ai.org/initiatives/ai-for-education/edtech-quality/benchmarks?benchmark=CDPK&cost=10&models=%5B%5D&pageView=table","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AI for Education Visual Reasoning - odd one out":{"url":"https://www.fab-ai.org/initiatives/ai-for-education/edtech-quality/benchmarks?benchmark=CDPK&cost=10&models=%5B%5D&pageView=table","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AI for Education Visual Reasoning - pattern completion (2d)":{"url":"https://www.fab-ai.org/initiatives/ai-for-education/edtech-quality/benchmarks?benchmark=CDPK&cost=10&models=%5B%5D&pageView=table","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AI for Education Visual Reasoning - pattern completion (linear)":{"url":"https://www.fab-ai.org/initiatives/ai-for-education/edtech-quality/benchmarks?benchmark=CDPK&cost=10&models=%5B%5D&pageView=table","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AI for Education Visual Reasoning - reasoning by analogy":{"url":"https://www.fab-ai.org/initiatives/ai-for-education/edtech-quality/benchmarks?benchmark=CDPK&cost=10&models=%5B%5D&pageView=table","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AI2D":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"sources":[{"name":"AI2D","url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)"},{"name":"OpenCompass Multimodal - AI2D","url":"https://rank.opencompass.org.cn","metric":"Score (%)"}],"random_baseline":25},"AIIQ Composite IQ":{"url":"https://benchmarklist.com/benchmarks/aiiq_composite_iq/","metric":"Composite IQ (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AILuminate Safety":{"url":"https://ailuminate.mlcommons.org/benchmarks/","metric":"Safety Grade (1-5)","theoretical_min":1.0,"theoretical_max":5.0},"AIM-Bench (inventory manager)":{"url":"https://benchmarklist.com/benchmarks/aim_bench_inventory_manager/","metric":"BG Avg. C (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AIMS":{"url":"https://benchmarklist.com/benchmarks/aims/","metric":"Average (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AIPatient Arena":{"url":"https://benchmarklist.com/benchmarks/aipatient_arena_ehr_grounded_evaluation_of_large_language_models_in_end_to_end_clinical_consultation_workflows/","metric":"Questioning Skills (QS) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AISI Cyber CTF":{"url":"https://arxiv.org/abs/2603.11214","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"AISI Cyber Cooling Tower 100M":{"url":"https://arxiv.org/abs/2603.11214","metric":"Avg Steps (/7)","theoretical_min":0.0,"theoretical_max":7.0},"AISI Cyber Cooling Tower 10M":{"url":"https://arxiv.org/abs/2603.11214","metric":"Avg Steps (/7)","theoretical_min":0.0,"theoretical_max":7.0},"AISI Cyber TLO 100M":{"url":"https://www.aisi.gov.uk/blog/our-evaluation-of-openais-gpt-5-5-cyber-capabilities","metric":"Avg Steps (/32)","theoretical_min":0.0,"theoretical_max":32.0},"AISI Cyber TLO 10M":{"url":"https://arxiv.org/abs/2603.11214","metric":"Avg Steps (/32)","theoretical_min":0.0,"theoretical_max":32.0},"AJ-Bench":{"url":"https://aj-bench.github.io/","metric":"Overall Avg@3 (%)","theoretical_min":0.0,"theoretical_max":100.0},"ALARB":{"url":"https://benchmarklist.com/benchmarks/alarb/","metric":"Facts+regulations correct (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ALE-Bench":{"url":"https://benchmarklist.com/benchmarks/ale_bench/","metric":"Performance (Self-Refine x1) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ALEM (Multi-Agent Coordination)":{"url":"https://alem-world.github.io/","metric":"Total Achievement % (medium)","theoretical_min":0.0,"theoretical_max":100.0},"ALL Bench LLM":{"url":"https://huggingface.co/datasets/youssef3146/ALL-Bench-Leaderboard","metric":"Average Numeric Benchmark Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"ALL Bench Multimodal":{"url":"https://huggingface.co/datasets/youssef3146/ALL-Bench-Leaderboard","metric":"Average Numeric VLM Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AMA-Bench":{"url":"https://huggingface.co/spaces/AMA-bench/AMA-bench-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AMA-Bench - Embodied AI":{"url":"https://huggingface.co/spaces/AMA-bench/AMA-bench-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AMA-Bench - Game":{"url":"https://huggingface.co/spaces/AMA-bench/AMA-bench-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AMA-Bench - Open-World QA":{"url":"https://huggingface.co/spaces/AMA-bench/AMA-bench-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AMA-Bench - Software":{"url":"https://huggingface.co/spaces/AMA-bench/AMA-bench-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AMA-Bench - Text2SQL":{"url":"https://huggingface.co/spaces/AMA-bench/AMA-bench-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AMA-Bench - Web":{"url":"https://huggingface.co/spaces/AMA-bench/AMA-bench-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"APEX v1":{"url":"https://www.mercor.com/apex/apex-v1-leaderboard/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"APEX v1 Big Law":{"url":"https://www.mercor.com/apex/apex-v1-leaderboard/big-law-associate/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"APEX v1 Consulting":{"url":"https://www.mercor.com/apex/apex-v1-leaderboard/consulting-associate/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"APEX v1 Investment Banking":{"url":"https://www.mercor.com/apex/apex-v1-leaderboard/investment-banking-analyst/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"APEX v1 Medicine (MD)":{"url":"https://www.mercor.com/apex/apex-v1-leaderboard/general-practitioner-md/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"APEX-Agents":{"url":"https://benchmarklist.com/benchmarks/apex_agents/","metric":"Mean Score (ReAct) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"APEX-Agents-AA":{"url":"https://benchmarklist.com/benchmarks/apex_agents_aa/","metric":"Pass@1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ARC Challenge (AI2)":{"url":"https://epoch.ai/benchmarks/arc-ai2","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":25},"ARC-AGI-1":{"url":"https://arcprize.org/leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"ARC-AGI-2":{"url":"https://arcprize.org/leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"ARC-AGI-2 Verified (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"ARC-AGI-3":{"url":"https://arcprize.org/leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"ARFBench":{"url":"https://huggingface.co/spaces/Datadog/ARFBench","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"ARFBench (Text Input)":{"url":"https://huggingface.co/spaces/Datadog/ARFBench","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"ASCIIBench":{"url":"https://www.asciibench.com/leaderboard","metric":"ELO Rating","theoretical_min":null,"theoretical_max":null},"ASCIIEval":{"url":"https://github.com/JiaQiSJTU/VisionInText","metric":"Macro-Avg Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"ASPI":{"url":"https://benchmarklist.com/benchmarks/aspi_seeking_ambiguity_clarification_amplifies_prompt_injection_vulnerability_in_llm_agents/","metric":"ASR exec_tool (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ATC Safety-Oriented Language Understanding Eval":{"url":"https://benchmarklist.com/benchmarks/atc_safety_eval/","metric":"Risk Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ATLAS":{"url":"https://benchmarklist.com/benchmarks/atlas/","metric":"ATLAScore (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ATRBench":{"url":"https://benchmarklist.com/benchmarks/atrbench/","metric":"TSAcc Default (%) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AV-Odyssey":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Action Prediction":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Action Recognition":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Action Sequencing":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Action Tracing":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Animal Recognition":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Audio 3D Angle Estimation":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Audio Distance Estimation":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Audio Time Estimation":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Audio-Visual Synchronization":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Bird Recognition":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Dance and Music Matching":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Eating Sound Recognition":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Film and Music Matching":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Gunshot Recognition":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Hallucination":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Hazard Recognition":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Instrument Recognition":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Intricacy":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Material Recognition":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Melody":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Meme Understanding":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Music Genre Classification":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Music Score Matching":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Music Sentiment Analysis":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Scene Recognition":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Singer Recognition":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Space":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Speech Sentiment Analysis":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Timbre":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Time":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Tone":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AV-Odyssey - Transportation Recognition":{"url":"https://huggingface.co/spaces/av-odyssey/AV_Odyssey_Bench_Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AcademiClaw":{"url":"https://benchmarklist.com/benchmarks/academiclaw_when_students_set_challenges_for_ai_agents/","metric":"Pass Rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ActTraitBench":{"url":"https://benchmarklist.com/benchmarks/acttraitbench_quantifying_the_knowledge_decision_gap_in_large_language_models_via_human_grounded_behavioral_validation/","metric":"G_KD (Global Knowledge-Decision Gap) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Active Evidence-Seeking and Diagnostic Reasoni":{"url":"https://benchmarklist.com/benchmarks/active_evidence_seeking_and_diagnostic_reasoning_in_large_language_models_for_clinical_decision_support/","metric":"Task 2: Active Seeking Exact Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AcuityBench":{"url":"https://benchmarklist.com/benchmarks/acuitybench/","metric":"QA Exact (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AfroBench":{"url":"https://huggingface.co/spaces/McGill-NLP/AfroBench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0,"sources":[{"name":"AfroBench","url":"https://huggingface.co/spaces/McGill-NLP/AfroBench","metric":"Average Score (%)"},{"name":"AfroBench-Lite","url":"https://benchmarklist.com/benchmarks/afrobench_lite/","metric":"Average score (self-reported)"}]},"Age of LLM":{"url":"https://benchmarklist.com/benchmarks/age_of_llm_a_strategic_1v1_benchmark_for_reasoning_diplomacy_and_reliability_of_large_language_models_under_fog_of_war/","metric":"Points per match (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Agent Arena":{"url":"https://arena.ai/leaderboard/agent","metric":"Net Improvement (%)","theoretical_min":0.0,"theoretical_max":100.0},"Agent Arena - Bash Recovery":{"url":"https://arena.ai/leaderboard/agent","metric":"Bash Recovery (%)","theoretical_min":0.0,"theoretical_max":100.0},"Agent Arena - Confirmed Success":{"url":"https://arena.ai/leaderboard/agent","metric":"Confirmed Success (%)","theoretical_min":0.0,"theoretical_max":100.0},"Agent Arena - Praise vs Complaint":{"url":"https://arena.ai/leaderboard/agent","metric":"Praise vs Complaint (%)","theoretical_min":0.0,"theoretical_max":100.0},"Agent Arena - Steerability":{"url":"https://arena.ai/leaderboard/agent","metric":"Steerability (%)","theoretical_min":0.0,"theoretical_max":100.0},"Agent Arena - Tool Hallucination":{"url":"https://arena.ai/leaderboard/agent","metric":"Tool Hallucination (%)","theoretical_min":0.0,"theoretical_max":100.0},"Agent Security League - Functional Correctness":{"url":"https://www.endorlabs.com/research/ai-code-security-benchmark","metric":"Functional Correctness (%)","theoretical_min":0.0,"theoretical_max":100.0},"Agent Security League - Security Correctness":{"url":"https://www.endorlabs.com/research/ai-code-security-benchmark","metric":"Security Correctness (%)","theoretical_min":0.0,"theoretical_max":100.0},"Agent-ValueBench":{"url":"https://benchmarklist.com/benchmarks/agent_valuebench/","metric":"Authority (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AgentBench FC":{"url":"https://github.com/THUDM/AgentBench","metric":"Avg Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"AgentBoard":{"url":"https://benchmarklist.com/benchmarks/agentboard/","metric":"Progress Rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AgentCIBench":{"url":"https://benchmarklist.com/benchmarks/agentcibench/","metric":"Leakage (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AgentClinic":{"url":"https://benchmarklist.com/benchmarks/agentclinic/","metric":"AgentClinic-MedQA diagnostic accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AgentCollabBench":{"url":"https://benchmarklist.com/benchmarks/agentcollabbench/","metric":"IDR (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AgentFloor":{"url":"https://benchmarklist.com/benchmarks/agentfloor/","metric":"TCR (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AgentIF":{"url":"https://benchmarklist.com/benchmarks/agentif/","metric":"Constraint Success Rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AgentLeak":{"url":"https://benchmarklist.com/benchmarks/agentleak/","metric":"Total Leak (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AgentTrap":{"url":"https://benchmarklist.com/benchmarks/agenttrap/","metric":"Attack success (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Agentic Skills Evaluation Framework":{"url":"https://benchmarklist.com/benchmarks/agentic_skills_evaluation_framework/","metric":"Overall Score w/ (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AgenticVBench":{"url":"https://agenticvbench.com/leaderboard","metric":"Average Success (%)","theoretical_min":0.0,"theoretical_max":100.0},"Agentick":{"url":"https://benchmarklist.com/benchmarks/agentick/","metric":"Oracle-Normalized Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Agents' Last Exam":{"url":"https://snorkel.ai/leaderboard/agents-last-exam/","metric":"Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"AgingBench":{"url":"https://benchmarklist.com/benchmarks/agingbench/","metric":"S1 kw_m HL (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AgroVG":{"url":"https://benchmarklist.com/benchmarks/agrovg/","metric":"All@.5 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AidanBench":{"url":"https://aidanbench.com/","metric":"Novel Answers","theoretical_min":null,"theoretical_max":null},"Aider Refactoring Benchmark":{"url":"https://benchmarklist.com/benchmarks/aider_refactoring_benchmark/","metric":"Percent completed correctly (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Aider polyglot coding leaderboard":{"url":"https://aider.chat/docs/leaderboards/","metric":"Pass rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Aikido CVE Rediscovery":{"url":"https://www.aikido.dev/blog/benchmarking-ai-models-known-cves","metric":"Recall (%)","theoretical_min":0.0,"theoretical_max":100.0},"Aikido CVE Rediscovery (pass@3)":{"url":"https://www.aikido.dev/blog/benchmarking-ai-models-known-cves","metric":"Recall (%)","theoretical_min":0.0,"theoretical_max":100.0},"AlpacaEval 1.0":{"url":"https://tatsu-lab.github.io/alpaca_eval/","metric":"Win Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"AlpacaEval 2.0":{"url":"https://tatsu-lab.github.io/alpaca_eval/","metric":"LC Win Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Alpha Arena (Nof1)":{"url":"https://nof1.ai/leaderboard","metric":"Return (%)","theoretical_min":null,"theoretical_max":null},"Ambig-DS":{"url":"https://benchmarklist.com/benchmarks/ambig_ds/","metric":"Full (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AmongAIs":{"url":"https://www.antimlabs.com/amongais","metric":"Score (50×Impostor + 10×Crewmate wins)","theoretical_min":null,"theoretical_max":null},"An Empirical Study of Proactive Coding Assista":{"url":"https://benchmarklist.com/benchmarks/an_empirical_study_of_proactive_coding_assistants_in_real_world_software_development/","metric":"Pass@1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Android Bench":{"url":"https://developer.android.com/bench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AndroidWorld":{"url":"https://github.com/google-research/android_world","metric":"Success Rate pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"Anthropic ECI (AECI)":{"url":"https://abstatisticalconsulting.substack.com/p/predicting-time-horizon-from-anthropics","metric":"AECI Score","theoretical_min":null,"theoretical_max":null},"AppWorld Challenge":{"url":"https://appworld.dev/leaderboard","metric":"Task Goal Completion (%)","theoretical_min":0.0,"theoretical_max":100.0},"AppWorld Normal":{"url":"https://appworld.dev/leaderboard","metric":"Task Goal Completion (%)","theoretical_min":0.0,"theoretical_max":100.0},"Appwrite Arena (With Skills)":{"url":"https://raw.githubusercontent.com/appwrite/arena/main/src/data/results-with-skills.json","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Appwrite Arena (Without Skills)":{"url":"https://raw.githubusercontent.com/appwrite/arena/main/src/data/results-without-skills.json","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"ArXivMath Mar-Apr 2026 (Anthropic)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"AraGen":{"url":"https://huggingface.co/spaces/inceptionai/Arabic-Leaderboards","metric":"3C3H Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Arabic Broad Leaderboard":{"url":"https://huggingface.co/spaces/silma-ai/Arabic-LLM-Broad-Leaderboard","metric":"Average Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"Arabic IFEval":{"url":"https://huggingface.co/spaces/inceptionai/Leaderboards","metric":"Arabic Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Arcadia CommonsenseQA":{"url":"https://huggingface.co/datasets/tau/commonsense_qa","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Arcadia MMMU Multiple Choice":{"url":"https://huggingface.co/datasets/MMMU/MMMU","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Arcadia MMMU Open Ended":{"url":"https://huggingface.co/datasets/MMMU/MMMU","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"ArchSIBench":{"url":"https://benchmarklist.com/benchmarks/archsibench_benchmarking_the_architectural_spatial_intelligence_of_vision_language_models/","metric":"Avg. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Are Agents Ready to Teach? A Multi-Stage Bench":{"url":"https://benchmarklist.com/benchmarks/are_agents_ready_to_teach_a_multi_stage_benchmark_for_real_world_teaching_workflows/","metric":"Eq. pass (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Are LLMs Ready to Assist Physicians? PhysAssis":{"url":"https://benchmarklist.com/benchmarks/are_llms_ready_to_assist_physicians_physassistbench_for_interactive_doctor_patient_ehr_assistance/","metric":"en_mrs (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Arena AI Code":{"url":"https://benchmarklist.com/benchmarks/arena_ai_code/","metric":"Arena ELO (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Arena AI Document":{"url":"https://benchmarklist.com/benchmarks/arena_ai_document/","metric":"Arena ELO (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Arena-Hard Creative Writing":{"url":"https://github.com/lmarena/arena-hard-auto","metric":"Win Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Arena-Hard v2":{"url":"https://github.com/lmarena/arena-hard-auto","metric":"Win Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Arena-Hard v2 (GPT-4.1 Judge)":{"url":"https://github.com/lmarena/arena-hard-auto","metric":"Win Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"ArtifactsBench":{"url":"https://benchmarklist.com/benchmarks/artifactsbench/","metric":"AVG (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ArtifactsBenchmark":{"url":"https://github.com/Tencent-Hunyuan/ArtifactsBenchmark","metric":"Average Score","theoretical_min":0.0,"theoretical_max":null},"Artificial Analysis Intelligence Index":{"url":"https://artificialanalysis.ai/evaluations/artificial-analysis-intelligence-index","metric":"Intelligence Index","theoretical_min":0.0,"theoretical_max":100.0},"Artificial Intelligence Quotient (AIQ) Benchmark":{"url":"https://benchmarklist.com/benchmarks/artificial_intelligence_quotient_aiq_benchmark/","metric":"Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ArxivMath":{"url":"https://benchmarklist.com/benchmarks/arxivmath/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ArxivMath (Fable/Mythos)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AssertLLM2":{"url":"https://benchmarklist.com/benchmarks/assertllm2/","metric":"Coverage Formal (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AsyncTool":{"url":"https://benchmarklist.com/benchmarks/asynctool/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AttuneBench":{"url":"https://benchmarklist.com/benchmarks/attunebench/","metric":"Composite (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AudioBench":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - AISHELL ASR ZH WER":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"WER (%) (lower is better)","theoretical_min":0.0,"theoretical_max":null},"AudioBench - Alpaca Audio":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - AudioCaps Captioning":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - AudioCaps QA":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - CN College Listen MCQ":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - Clotho AQA":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - CoVoST2 EN-ID BLEU":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - CoVoST2 EN-TA BLEU":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - CoVoST2 EN-ZH BLEU":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - CoVoST2 ID-EN BLEU":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - CoVoST2 TA-EN BLEU":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - CoVoST2 ZH-EN BLEU":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - Common Voice 15 EN WER":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"WER (%) (lower is better)","theoretical_min":0.0,"theoretical_max":null},"AudioBench - DREAM TTS MCQ":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - Earnings21 WER":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"WER (%) (lower is better)","theoretical_min":0.0,"theoretical_max":null},"AudioBench - Earnings22 WER":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"WER (%) (lower is better)","theoretical_min":0.0,"theoretical_max":null},"AudioBench - GigaSpeech WER":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"WER (%) (lower is better)","theoretical_min":0.0,"theoretical_max":null},"AudioBench - IEMOCAP Emotion":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - IEMOCAP Gender":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - IMDA AR Dialogue":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - IMDA AR Sentence":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - IMDA GR Dialogue":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - IMDA GR Sentence":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - IMDA Part 1 ASR WER":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"WER (%) (lower is better)","theoretical_min":0.0,"theoretical_max":null},"AudioBench - IMDA Part 2 ASR WER":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"WER (%) (lower is better)","theoretical_min":0.0,"theoretical_max":null},"AudioBench - IMDA Part 3 ASR WER":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"WER (%) (lower is better)","theoretical_min":0.0,"theoretical_max":null},"AudioBench - IMDA Part 3 Dialogue Summarization Human":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - IMDA Part 3 SQA Human":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - IMDA Part 4 ASR WER":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"WER (%) (lower is better)","theoretical_min":0.0,"theoretical_max":null},"AudioBench - IMDA Part 4 Dialogue Summarization Human":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - IMDA Part 4 SQA Human":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - IMDA Part 5 ASR WER":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"WER (%) (lower is better)","theoretical_min":0.0,"theoretical_max":null},"AudioBench - IMDA Part 5 Dialogue Summarization Human":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - IMDA Part 5 SQA Human":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - IMDA Part 6 ASR WER":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"WER (%) (lower is better)","theoretical_min":0.0,"theoretical_max":null},"AudioBench - IMDA Part 6 Dialogue Summarization Human":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - IMDA Part 6 SQA Human":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - LibriSpeech Clean WER":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"WER (%) (lower is better)","theoretical_min":0.0,"theoretical_max":null},"AudioBench - LibriSpeech Other WER":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"WER (%) (lower is better)","theoretical_min":0.0,"theoretical_max":null},"AudioBench - MELD Emotion":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - MELD Sentiment":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - MMAU Mini":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - MMAU Music":{"url":"https://huggingface.co/spaces/AudioLLMs/AudioBench-Leaderboard-Extend","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - MMAU Sound":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - MMAU Speech":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - MuChoMusic":{"url":"https://huggingface.co/spaces/AudioLLMs/AudioBench-Leaderboard-Extend","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - OpenHermes Audio":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - Public SG Speech QA":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - SEAME Mandarin WER":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"WER (%) (lower is better)","theoretical_min":0.0,"theoretical_max":null},"AudioBench - SEAME Singapore English WER":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"WER (%) (lower is better)","theoretical_min":0.0,"theoretical_max":null},"AudioBench - SLUE P2 SQA5":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - Spoken SQuAD":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - TED-LIUM 3 Long Form WER":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"WER (%) (lower is better)","theoretical_min":0.0,"theoretical_max":null},"AudioBench - TED-LIUM 3 WER":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"WER (%) (lower is better)","theoretical_min":0.0,"theoretical_max":null},"AudioBench - VoxCeleb Accent":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - VoxCeleb Gender":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - WavCaps Captioning":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - WavCaps QA":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - YouTube ASR Batch 1 WER":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"WER (%) (lower is better)","theoretical_min":0.0,"theoretical_max":null},"AudioBench - YouTube ASR Batch 2 WER":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"WER (%) (lower is better)","theoretical_min":0.0,"theoretical_max":null},"AudioBench - YouTube PQA":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - YouTube SDS":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench - YouTube SQA":{"url":"https://github.com/AudioLLMs/AudioBench","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioBench Music":{"url":"https://huggingface.co/spaces/AudioLLMs/AudioBench-Leaderboard-Extend","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"AudioMC":{"url":"https://benchmarklist.com/benchmarks/audiomc/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AutoBench":{"url":"https://autobench.org/","metric":"AutoBench Rank","theoretical_min":1.0,"theoretical_max":5.0},"AutoLab":{"url":"https://benchmarklist.com/benchmarks/autolab/","metric":"Overall Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AutoLogi":{"url":"https://benchmarklist.com/benchmarks/autologi/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AutoMedBench":{"url":"https://benchmarklist.com/benchmarks/automedbench/","metric":"Average Overall Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"AutomataBench":{"url":"https://www.automatabench.com/","metric":"Weighted pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"AutomationBench":{"url":"https://zapier.com/benchmarks","metric":"Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"AutomationBench (Fable/Mythos)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BABILong (NIAH)":{"url":"https://huggingface.co/spaces/RMT-team/babilong","metric":"Avg Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"BALROG BabaIsAI (LLM)":{"url":"https://balrogai.com/","metric":"Progress (%)","theoretical_min":0.0,"theoretical_max":100.0},"BALROG BabaIsAI (VLM)":{"url":"https://balrogai.com/","metric":"Progress (%)","theoretical_min":0.0,"theoretical_max":100.0},"BALROG BabyAI (LLM)":{"url":"https://balrogai.com/","metric":"Progress (%)","theoretical_min":0.0,"theoretical_max":100.0},"BALROG BabyAI (VLM)":{"url":"https://balrogai.com/","metric":"Progress (%)","theoretical_min":0.0,"theoretical_max":100.0},"BALROG Crafter (LLM)":{"url":"https://balrogai.com/","metric":"Progress (%)","theoretical_min":0.0,"theoretical_max":100.0},"BALROG Crafter (VLM)":{"url":"https://balrogai.com/","metric":"Progress (%)","theoretical_min":0.0,"theoretical_max":100.0},"BALROG MiniHack (LLM)":{"url":"https://balrogai.com/","metric":"Progress (%)","theoretical_min":0.0,"theoretical_max":100.0},"BALROG MiniHack (VLM)":{"url":"https://balrogai.com/","metric":"Progress (%)","theoretical_min":0.0,"theoretical_max":100.0},"BALROG NetHack (LLM)":{"url":"https://balrogai.com/","metric":"Progress (%)","theoretical_min":0.0,"theoretical_max":100.0},"BALROG NetHack (VLM)":{"url":"https://balrogai.com/","metric":"Progress (%)","theoretical_min":0.0,"theoretical_max":100.0},"BALROG TextWorld (LLM)":{"url":"https://balrogai.com/","metric":"Progress (%)","theoretical_min":0.0,"theoretical_max":100.0},"BBEH":{"url":"https://github.com/google-deepmind/bbeh","metric":"Harmonic Mean (%)","theoretical_min":0.0,"theoretical_max":100.0},"BBH":{"url":"https://benchmarklist.com/benchmarks/bbh_epoch/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"BELLS":{"url":"https://centresecuriteia.github.io/bells_leaderboard/leaderboard.html","metric":"BELLS Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BIG-Bench Extra Hard":{"url":"https://benchmarklist.com/benchmarks/big_bench_extra_hard/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"BIM-Edit":{"url":"https://benchmarklist.com/benchmarks/bim_edit/","metric":"Final (self-reported)","theoretical_min":0.0,"theoretical_max":null},"BIRD-CRITIC":{"url":"https://github.com/bird-bench/BIRD-CRITIC-1","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"BIRD-Interact (a-Interact)":{"url":"https://github.com/bird-bench/BIRD-Interact","metric":"Normalized Reward","theoretical_min":null,"theoretical_max":null},"BIRD-Interact (c-Interact)":{"url":"https://github.com/bird-bench/BIRD-Interact","metric":"Normalized Reward","theoretical_min":null,"theoretical_max":null},"BIRD-SQL":{"url":"https://bird-bench.github.io/","metric":"Execution Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"BLUEX v2":{"url":"https://benchmarklist.com/benchmarks/bluex_v2/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"BLXBench":{"url":"https://benchmarklist.com/benchmarks/blxbench/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"BRIDGE Medical Leaderboard":{"url":"https://huggingface.co/spaces/YLab-Open/BRIDGE-Medical-Leaderboard","metric":"Average Performance (%)","theoretical_min":0.0,"theoretical_max":100.0},"BRIDGE Medical Leaderboard - CoT":{"url":"https://huggingface.co/spaces/YLab-Open/BRIDGE-Medical-Leaderboard","metric":"Average Performance (%)","theoretical_min":0.0,"theoretical_max":100.0},"BRIDGE Medical Leaderboard - Few-Shot":{"url":"https://huggingface.co/spaces/YLab-Open/BRIDGE-Medical-Leaderboard","metric":"Average Performance (%)","theoretical_min":0.0,"theoretical_max":100.0},"BRIDGE Medical Leaderboard - Zero-Shot":{"url":"https://huggingface.co/spaces/YLab-Open/BRIDGE-Medical-Leaderboard","metric":"Average Performance (%)","theoretical_min":0.0,"theoretical_max":100.0},"BTZSC":{"url":"https://huggingface.co/spaces/btzsc/btzsc-leaderboard","metric":"Macro-F1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"BTZSC - Emotion":{"url":"https://huggingface.co/spaces/btzsc/btzsc-leaderboard","metric":"Macro-F1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"BTZSC - Intent":{"url":"https://huggingface.co/spaces/btzsc/btzsc-leaderboard","metric":"Macro-F1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"BTZSC - LLM":{"url":"https://huggingface.co/spaces/btzsc/btzsc-leaderboard","metric":"Macro-F1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"BTZSC - Sentiment":{"url":"https://huggingface.co/spaces/btzsc/btzsc-leaderboard","metric":"Macro-F1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"BTZSC - Topic":{"url":"https://huggingface.co/spaces/btzsc/btzsc-leaderboard","metric":"Macro-F1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"BabyVision":{"url":"https://unipat.ai/benchmarks/BabyVision","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BacktestBench":{"url":"https://benchmarklist.com/benchmarks/backtestbench/","metric":"Overall Accuracy (OA) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Balrog":{"url":"https://benchmarklist.com/benchmarks/balrog/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"BehaviorBench":{"url":"https://benchmarklist.com/benchmarks/behaviorbench_benchmarking_foundation_models_for_behavioral_science_tasks/","metric":"Game Behav. Sim. (W ↓) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"BenCzechMark":{"url":"https://huggingface.co/spaces/CZLC/BenCzechMark","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BenGER":{"url":"https://arxiv.org/html/2605.28183v1","metric":"Aggregate Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"BenchBench":{"url":"https://huggingface.co/spaces/per/BenchBench","metric":"Aggregate Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BenchLM":{"url":"https://benchlm.ai/","metric":"Overall Score","theoretical_min":0.0,"theoretical_max":100.0},"BenchTable":{"url":"https://dubesor.de/benchtable","metric":"Total Score (%)","theoretical_min":null,"theoretical_max":100.0},"Benchmarking Large Language Models for Graphem":{"url":"https://benchmarklist.com/benchmarks/benchmarking_large_language_models_for_grapheme_to_phoneme_conversion_a_japanese_case_study/","metric":"Direct (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Benchmarks.bio - BioSecBench-Refusal":{"url":"https://benchmarks.bio/security","metric":"Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Benchmarks.bio - BioSecBench-Surveillance":{"url":"https://benchmarks.bio/surveillance","metric":"Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Benchmarks.bio - EpiBench":{"url":"https://benchmarks.bio/","metric":"Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Benchmarks.bio - SpatialBench-Long":{"url":"https://benchmarks.bio/spatial-long","metric":"Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Benchmarks.bio - TxBench-PP":{"url":"https://benchmarks.bio/?bench=txbench","metric":"Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Benchmarks.bio - VariantBench":{"url":"https://benchmarks.bio/variant","metric":"Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Benchmarks.bio - scBench-Long":{"url":"https://benchmarks.bio/scbench-long","metric":"Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Beyond Binary":{"url":"https://benchmarklist.com/benchmarks/beyond_binary_reframing_gui_critique_as_continuous_semantic_alignment/","metric":"NDCG@All (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Beyond Continuity":{"url":"https://benchmarklist.com/benchmarks/beyond_continuity_challenges_of_context_switching_in_multi_turn_dialogue_with_llms/","metric":"F1 Pivot (TopiOCQA) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Beyond Function Calling":{"url":"https://benchmarklist.com/benchmarks/beyond_function_calling_benchmarking_tool_using_agents_under_tool_environment_unreliability/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Beyond Pass Rate":{"url":"https://arxiv.org/html/2606.08840v1","metric":"Correctness (%)","theoretical_min":0.0,"theoretical_max":100.0},"Beyond the All-in-One Agent":{"url":"https://benchmarklist.com/benchmarks/beyond_the_all_in_one_agent_benchmarking_role_specialized_multi_agent_collaboration_in_enterprise_workflows/","metric":"avg. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"BiGGen-Bench":{"url":"https://huggingface.co/spaces/prometheus-eval/BiGGen-Bench-Leaderboard","metric":"Average Score (1-5)","theoretical_min":1.0,"theoretical_max":5.0},"Big Code Memorization - HumanEval pass@1":{"url":"https://huggingface.co/spaces/wzxii/Memorization-or-Generation-of-Big-Code-Models-Leaderboard","metric":"HumanEval pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"Big Code Memorization - HumanEval pass@50":{"url":"https://huggingface.co/spaces/wzxii/Memorization-or-Generation-of-Big-Code-Models-Leaderboard","metric":"HumanEval pass@50 (%)","theoretical_min":0.0,"theoretical_max":100.0},"Big Code Memorization - HumanEval-ET pass@1":{"url":"https://huggingface.co/spaces/wzxii/Memorization-or-Generation-of-Big-Code-Models-Leaderboard","metric":"HumanEval-ET pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"Big Code Memorization - HumanEval-ET pass@50":{"url":"https://huggingface.co/spaces/wzxii/Memorization-or-Generation-of-Big-Code-Models-Leaderboard","metric":"HumanEval-ET pass@50 (%)","theoretical_min":0.0,"theoretical_max":100.0},"Big-Bench Hard":{"url":"https://epoch.ai/benchmarks/bbh","metric":"Average (%)","theoretical_min":0.0,"theoretical_max":100.0},"BigCode Models Leaderboard":{"url":"https://huggingface.co/spaces/bigcode/bigcode-models-leaderboard","metric":"HumanEval Python Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"BigCodeBench":{"url":"https://bigcode-bench.github.io/","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"BigCodeBench-Hard":{"url":"https://benchmarklist.com/benchmarks/bigcodebench_hard/","metric":"Instruct pass@1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"BilliardPhys-Bench":{"url":"https://benchmarklist.com/benchmarks/billiardphys_bench/","metric":"Total (Weighted) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"BinaryAudit":{"url":"https://quesma.com/benchmarks/binaryaudit/","metric":"Avg Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"BioDesignBench":{"url":"https://huggingface.co/spaces/RomeroLab-Duke/BioDesignBench-Leaderboard","metric":"Overall Score","theoretical_min":0.0,"theoretical_max":100.0},"BioMysteryBench (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BioMysteryBench Human Difficult (Fable/Mythos)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BioMysteryBench Human Solvable (Fable/Mythos)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BioMysteryBench Human-Difficult":{"url":"https://benchmarklist.com/benchmarks/biomysterybench_human_difficult/","metric":"Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"BioMysteryBench Human-Solvable":{"url":"https://benchmarklist.com/benchmarks/biomysterybench_human_solvable/","metric":"Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"BioMysteryBench Verified - Human Difficult (Anthropic)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BioMysteryBench Verified - Human Solvable (Anthropic)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BioPipelineBench (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BioPipelineBench Verified":{"url":"https://benchmarklist.com/benchmarks/biopipelinebench_verified/","metric":"Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"BioPipelineBench Verified (Anthropic)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BizFinBench":{"url":"https://github.com/HiThink-Research/BizFinBench","metric":"Average Score","theoretical_min":0.0,"theoretical_max":100.0},"BlueBench":{"url":"https://huggingface.co/spaces/ibm-research/bluebench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BlueBench - Bias":{"url":"https://huggingface.co/spaces/ibm-research/bluebench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BlueBench - Chatbot Abilities":{"url":"https://huggingface.co/spaces/ibm-research/bluebench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BlueBench - Entity Extraction":{"url":"https://huggingface.co/spaces/ibm-research/bluebench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BlueBench - Knowledge":{"url":"https://huggingface.co/spaces/ibm-research/bluebench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BlueBench - Legal":{"url":"https://huggingface.co/spaces/ibm-research/bluebench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BlueBench - News Classification":{"url":"https://huggingface.co/spaces/ibm-research/bluebench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BlueBench - Product Help":{"url":"https://huggingface.co/spaces/ibm-research/bluebench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BlueBench - QA Finance":{"url":"https://huggingface.co/spaces/ibm-research/bluebench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BlueBench - RAG General":{"url":"https://huggingface.co/spaces/ibm-research/bluebench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BlueBench - Reasoning":{"url":"https://huggingface.co/spaces/ibm-research/bluebench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BlueBench - Safety":{"url":"https://huggingface.co/spaces/ibm-research/bluebench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BlueBench - Summarization":{"url":"https://huggingface.co/spaces/ibm-research/bluebench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BlueBench - Translation":{"url":"https://huggingface.co/spaces/ibm-research/bluebench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Blueprint-Bench":{"url":"https://andonlabs.com/evals/blueprint-bench","metric":"Similarity Score","theoretical_min":0.0,"theoretical_max":1.0},"Blueprint-Bench 2":{"url":"https://andonlabs.com/evals/blueprint-bench-2","metric":"Connectivity Similarity Score","theoretical_min":0.0,"theoretical_max":1.0},"Blueprint-Bench 2 (Fable/Mythos)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Boiling the Frog":{"url":"https://benchmarklist.com/benchmarks/boiling_the_frog/","metric":"Strict ASR (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Bongard in Wonderland":{"url":"https://ml-research.github.io/bongard-in-wonderland/","metric":"Problems Solved (out of 100)","theoretical_min":0.0,"theoretical_max":100.0},"BoolQ":{"url":"https://epoch.ai/benchmarks/eci","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":50},"Bosses, Kings, and the Commons":{"url":"https://benchmarklist.com/benchmarks/bosses_kings_and_the_commons_cooperation_under_power_asymmetry_in_llm_societies/","metric":"Survival Rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"BoxPwnr CTF Bench":{"url":"https://github.com/0ca/BoxPwnr-Traces","metric":"Average Platform Completion (%)","theoretical_min":0.0,"theoretical_max":100.0},"BridgeBench Debugging":{"url":"https://www.bridgebench.ai/debugging","metric":"Confidence-weighted Elo","theoretical_min":null,"theoretical_max":null},"BrowseComp (Fable/Mythos Multi-Agent)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BrowseComp (Fable/Mythos Single-Agent)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"BrowserGym MiniWoB":{"url":"https://huggingface.co/spaces/ServiceNow/browsergym-leaderboard","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"BrowserGym WorkArena L1":{"url":"https://huggingface.co/spaces/ServiceNow/browsergym-leaderboard","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"BrowserGym WorkArena L2":{"url":"https://huggingface.co/spaces/ServiceNow/browsergym-leaderboard","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Bullshit Benchmark":{"url":"https://petergpt.github.io/bullshit-benchmark/viewer/index.html","metric":"BS Detection Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Business Utility Eval":{"url":"https://deepsense.ai/blog/business-utility-eval-june-2026-update/","metric":"Business Utility (%)","theoretical_min":0.0,"theoretical_max":100.0},"Buyout Game (Lechmazur)":{"url":"https://github.com/lechmazur/buyout_game","metric":"Bradley-Terry Rating","theoretical_min":null,"theoretical_max":null},"BuzzBench (Humour)":{"url":"https://eqbench.com/buzzbench.html","metric":"Humour Score (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"C3-Bench":{"url":"https://benchmarklist.com/benchmarks/c3_bench_a_context_aware_change_captioning_benchmark/","metric":"Aggregation (self-reported)","theoretical_min":0.0,"theoretical_max":null},"C4 Benchmark":{"url":"https://www.aviro.ai/writeups/c4-ebla","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"C4STYLI":{"url":"https://benchmarklist.com/benchmarks/c4styli/","metric":"Accuracy (C4Styli-T) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CADBench":{"url":"https://benchmarklist.com/benchmarks/cadbench_a_multimodal_benchmark_for_ai_assisted_cad_program_generation/","metric":"IoU Aggregate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CADGenBench":{"url":"https://huggingface.co/spaces/HuggingAI4Engineering/CADGenBench","metric":"Aggregate CAD Score","theoretical_min":0.0,"theoretical_max":1.0},"CAIA - Pass@1 (With Tools)":{"url":"https://huggingface.co/spaces/cyberco/CAIA_0927_Leaderboard","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"CAIA - Pass@1 (Without Tools)":{"url":"https://huggingface.co/spaces/cyberco/CAIA_0927_Leaderboard","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"CAIS Risk Index":{"url":"https://benchmarklist.com/benchmarks/cais_risk_index/","metric":"Risk Index (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CAIS Text Capabilities Index":{"url":"https://benchmarklist.com/benchmarks/cais_text_capabilities_index/","metric":"Text Capabilities Index (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CAIS Vision Capabilities Index":{"url":"https://benchmarklist.com/benchmarks/cais_vision_capabilities_index/","metric":"Vision Capabilities Index (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CAM-Bench":{"url":"https://benchmarklist.com/benchmarks/cam_bench/","metric":"Pass@32 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CAR-bench":{"url":"https://car-bench.github.io/car-bench/leaderboard.html","metric":"Avg Pass^3 (%)","theoretical_min":0.0,"theoretical_max":100.0},"CC-OCR V2":{"url":"https://benchmarklist.com/benchmarks/cc_ocr_v2/","metric":"Average (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CCBench":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"CEO-Bench":{"url":"https://benchmarklist.com/benchmarks/ceo_bench_can_agents_play_the_long_game/","metric":"Best run API cost (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CFAgentBench":{"url":"https://arxiv.org/html/2606.22000v1","metric":"pass^1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"CFMME":{"url":"https://benchmarklist.com/benchmarks/cfmme/","metric":"Average (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CHI-Bench":{"url":"https://actava.ai/benchmarks/leaderboards","metric":"Overall Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"CLAW-Eval":{"url":"https://claw-eval.github.io/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"CLBench":{"url":"https://www.clbench.com/","metric":"Solving Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"CLEAR":{"url":"https://benchmarklist.com/benchmarks/clear_revealing_how_noise_and_ambiguity_degrade_reliability_in_llms_for_medicine/","metric":"Version (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CLEM AdventureGame":{"url":"https://huggingface.co/spaces/colab-potsdam/clem-leaderboard","metric":"Game Clemscore (%)","theoretical_min":0.0,"theoretical_max":100.0},"CLEM Clean Up":{"url":"https://huggingface.co/spaces/colab-potsdam/clem-leaderboard","metric":"Game Clemscore (%)","theoretical_min":0.0,"theoretical_max":100.0},"CLEM Clemscore":{"url":"https://huggingface.co/spaces/colab-potsdam/clem-leaderboard","metric":"Clemscore (%)","theoretical_min":0.0,"theoretical_max":100.0},"CLEM Codenames":{"url":"https://huggingface.co/spaces/colab-potsdam/clem-leaderboard","metric":"Game Clemscore (%)","theoretical_min":0.0,"theoretical_max":100.0},"CLEM Deal or No Deal":{"url":"https://huggingface.co/spaces/colab-potsdam/clem-leaderboard","metric":"Game Clemscore (%)","theoretical_min":0.0,"theoretical_max":100.0},"CLEM GuessWhat":{"url":"https://huggingface.co/spaces/colab-potsdam/clem-leaderboard","metric":"Game Clemscore (%)","theoretical_min":0.0,"theoretical_max":100.0},"CLEM Hot Air Balloon":{"url":"https://huggingface.co/spaces/colab-potsdam/clem-leaderboard","metric":"Game Clemscore (%)","theoretical_min":0.0,"theoretical_max":100.0},"CLEM ImageGame":{"url":"https://huggingface.co/spaces/colab-potsdam/clem-leaderboard","metric":"Game Clemscore (%)","theoretical_min":0.0,"theoretical_max":100.0},"CLEM MatchIt ASCII":{"url":"https://huggingface.co/spaces/colab-potsdam/clem-leaderboard","metric":"Game Clemscore (%)","theoretical_min":0.0,"theoretical_max":100.0},"CLEM PrivateShared":{"url":"https://huggingface.co/spaces/colab-potsdam/clem-leaderboard","metric":"Game Clemscore (%)","theoretical_min":0.0,"theoretical_max":100.0},"CLEM ReferenceGame":{"url":"https://huggingface.co/spaces/colab-potsdam/clem-leaderboard","metric":"Game Clemscore (%)","theoretical_min":0.0,"theoretical_max":100.0},"CLEM Taboo":{"url":"https://huggingface.co/spaces/colab-potsdam/clem-leaderboard","metric":"Game Clemscore (%)","theoretical_min":0.0,"theoretical_max":100.0},"CLEM TextMapWorld":{"url":"https://huggingface.co/spaces/colab-potsdam/clem-leaderboard","metric":"Game Clemscore (%)","theoretical_min":0.0,"theoretical_max":100.0},"CLEM TextMapWorld GraphReasoning":{"url":"https://huggingface.co/spaces/colab-potsdam/clem-leaderboard","metric":"Game Clemscore (%)","theoretical_min":0.0,"theoretical_max":100.0},"CLEM TextMapWorld SpecificRoom":{"url":"https://huggingface.co/spaces/colab-potsdam/clem-leaderboard","metric":"Game Clemscore (%)","theoretical_min":0.0,"theoretical_max":100.0},"CLEM Wordle":{"url":"https://huggingface.co/spaces/colab-potsdam/clem-leaderboard","metric":"Game Clemscore (%)","theoretical_min":0.0,"theoretical_max":100.0},"CLEM Wordle with Clue":{"url":"https://huggingface.co/spaces/colab-potsdam/clem-leaderboard","metric":"Game Clemscore (%)","theoretical_min":0.0,"theoretical_max":100.0},"CLEM Wordle with Critic":{"url":"https://huggingface.co/spaces/colab-potsdam/clem-leaderboard","metric":"Game Clemscore (%)","theoretical_min":0.0,"theoretical_max":100.0},"CMMLU":{"url":"https://github.com/haonan-li/CMMLU","metric":"5-shot Avg Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":25},"CMMMU":{"url":"https://cmmmu-benchmark.github.io/","metric":"Test Overall (%)","theoretical_min":0.0,"theoretical_max":100.0},"CMMMU (Validation)":{"url":"https://cmmmu-benchmark.github.io/","metric":"Validation Overall (%)","theoretical_min":0.0,"theoretical_max":100.0},"CMT-Benchmark":{"url":"https://arxiv.org/abs/2510.05228","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"CPCD-Bench":{"url":"https://benchmarklist.com/benchmarks/cpcd_bench/","metric":"SR (LLM-as-a-Judge) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CPTU Bench":{"url":"https://huggingface.co/spaces/speakleash/cptu_bench","metric":"Average Score (1-5)","theoretical_min":1.0,"theoretical_max":5.0},"CRM LLM Leaderboard":{"url":"https://huggingface.co/spaces/Salesforce/crm_llm_leaderboard","metric":"CRM Accuracy (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"CRMArena - BRI":{"url":"https://huggingface.co/spaces/Salesforce/CRMArena-Leaderboard/resolve/main/crmarena_results/all_results.csv","metric":"BRI Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"CRMArena - HTU":{"url":"https://huggingface.co/spaces/Salesforce/CRMArena-Leaderboard/resolve/main/crmarena_results/all_results.csv","metric":"HTU Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"CRMArena - KQA":{"url":"https://huggingface.co/spaces/Salesforce/CRMArena-Leaderboard/resolve/main/crmarena_results/all_results.csv","metric":"KQA Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"CRMArena - MTA":{"url":"https://huggingface.co/spaces/Salesforce/CRMArena-Leaderboard/resolve/main/crmarena_results/all_results.csv","metric":"MTA Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"CRMArena - NCR":{"url":"https://huggingface.co/spaces/Salesforce/CRMArena-Leaderboard/resolve/main/crmarena_results/all_results.csv","metric":"NCR Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"CRMArena - NED":{"url":"https://huggingface.co/spaces/Salesforce/CRMArena-Leaderboard/resolve/main/crmarena_results/all_results.csv","metric":"NED Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"CRMArena - Overall":{"url":"https://huggingface.co/spaces/Salesforce/CRMArena-Leaderboard/resolve/main/crmarena_results/all_results.csv","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"CRMArena - PVI":{"url":"https://huggingface.co/spaces/Salesforce/CRMArena-Leaderboard/resolve/main/crmarena_results/all_results.csv","metric":"PVI Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"CRMArena - TCU":{"url":"https://huggingface.co/spaces/Salesforce/CRMArena-Leaderboard/resolve/main/crmarena_results/all_results.csv","metric":"TCU Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"CRMArena - TII":{"url":"https://huggingface.co/spaces/Salesforce/CRMArena-Leaderboard/resolve/main/crmarena_results/all_results.csv","metric":"TII Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"CRUST-bench":{"url":"https://crust-bench.github.io/","metric":"Pass@1 Test Success (%)","theoretical_min":0.0,"theoretical_max":100.0},"CRUXEval":{"url":"https://crux-eval.github.io/leaderboard.html","metric":"Output Prediction pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"CTFBench":{"url":"https://benchmarklist.com/benchmarks/ctfbench/","metric":"Vulnerability Detection Rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CTI-REALM":{"url":"https://arxiv.org/abs/2603.13517","metric":"Normalized Reward","theoretical_min":0.0,"theoretical_max":1.0},"CUDABeaver":{"url":"https://benchmarklist.com/benchmarks/cudabeaver_benchmarking_llm_based_automated_cuda_debugging/","metric":"pass@5 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CULTURE-MT":{"url":"https://benchmarklist.com/benchmarks/culture_mt/","metric":"Ineff. (%) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CUSP Science":{"url":"https://seanwu25.github.io/CUSP-Science/","metric":"FRQ Pass (%)","theoretical_min":0.0,"theoretical_max":100.0},"CVE-Bench":{"url":"https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf","metric":"pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"CVerifBench":{"url":"https://benchmarklist.com/benchmarks/cverifbench_or_unstated_the_evaluation_framework_of_500_c_verification_tasks/","metric":"Total (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CZ-EVAL - Analytical":{"url":"https://huggingface.co/spaces/hynky/CZ-EVAL","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"CZ-EVAL - Critical Thinking":{"url":"https://huggingface.co/spaces/hynky/CZ-EVAL","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"CZ-EVAL - Culture":{"url":"https://huggingface.co/spaces/hynky/CZ-EVAL","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"CZ-EVAL - Klokan (Math)":{"url":"https://huggingface.co/spaces/hynky/CZ-EVAL","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"CZ-EVAL - Verbal":{"url":"https://huggingface.co/spaces/hynky/CZ-EVAL","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"CadEval":{"url":"https://benchmarklist.com/benchmarks/cadeval/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CalBench":{"url":"https://benchmarklist.com/benchmarks/calbench_evaluating_coordination_privacy_trade_offs_in_multi_agent_llms/","metric":"Excess (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Can Agents Price a Reaction? Evaluating LLMs o":{"url":"https://benchmarklist.com/benchmarks/can_agents_price_a_reaction_evaluating_llms_on_chemical_cost_reasoning/","metric":"C.10 Clean (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Can LLM Agents Respond to Disasters? Benchmark":{"url":"https://benchmarklist.com/benchmarks/can_llm_agents_respond_to_disasters_benchmarking_heterogeneous_geospatial_reasoning_in_emergency_operations/","metric":"AVG (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Can LLMs Falsify?":{"url":"https://falsifiers.github.io/","metric":"Counterexample Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Can Large Language Models Handle Discourse Par":{"url":"https://benchmarklist.com/benchmarks/can_large_language_models_handle_discourse_particles_a_case_study_of_colloquial_malay/","metric":"Overall Avg (MS) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CanAiCode":{"url":"https://huggingface.co/spaces/mike-ravkine/can-ai-code-results","metric":"Junior-v2 Python Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"CapArena-Auto":{"url":"https://huggingface.co/datasets/yan111222/results","metric":"CapArena-Auto Score (avg)","theoretical_min":-100.0,"theoretical_max":100.0},"CapArena-Auto vs CogVLM-19B":{"url":"https://huggingface.co/datasets/yan111222/results","metric":"CapArena-Auto Score (vs CogVLM-19B)","theoretical_min":-100.0,"theoretical_max":100.0},"CapArena-Auto vs GPT-4o":{"url":"https://huggingface.co/datasets/yan111222/results","metric":"CapArena-Auto Score (vs GPT-4o)","theoretical_min":-100.0,"theoretical_max":100.0},"CapArena-Auto vs MiniCPM-8B":{"url":"https://huggingface.co/datasets/yan111222/results","metric":"CapArena-Auto Score (vs MiniCPM-8B)","theoretical_min":-100.0,"theoretical_max":100.0},"CarWashBench":{"url":"https://carwashbench.github.io/CarWashBench/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"CardioLens":{"url":"https://benchmarklist.com/benchmarks/cardiolens/","metric":"F1 Score (Random) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Carlini Applied LLM Benchmark":{"url":"https://github.com/carlini/yet-another-applied-llm-benchmark","metric":"Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Causal Sensitivity Score (CSS)":{"url":"https://benchmarklist.com/benchmarks/causal_sensitivity_score_css/","metric":"CSS (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ChLogic":{"url":"https://benchmarklist.com/benchmarks/chlogic/","metric":"English (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Chain-of-Procedure":{"url":"https://benchmarklist.com/benchmarks/chain_of_procedure_hierarchical_visual_language_reasoning_for_procedural_qa/","metric":"Zero-Shot (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ChaosBench-Logic v2":{"url":"https://benchmarklist.com/benchmarks/chaosbench_logic_v2_evaluating_llm_logical_reasoning_over_dynamical_systems_at_scale/","metric":"MCC (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CharXiv Reasoning (Anthropic No Tools)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"CharXiv Reasoning (Anthropic Tools)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"CharXiv Reasoning (Fable/Mythos No Tools)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"CharXiv Reasoning (Fable/Mythos Tools)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"CharXiv Reasoning No Tools (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"CharXiv Reasoning With Cropping (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"CharXiv-R":{"url":"https://benchmarklist.com/benchmarks/charxiv_r/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ChartFI":{"url":"https://benchmarklist.com/benchmarks/chartfi_benchmarking_faithfulness_and_insightfulness_of_chart_descriptions_from_multimodal_large_language_models/","metric":"Acuity (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ChartMuseum":{"url":"https://chartmuseum-leaderboard.github.io/","metric":"Overall Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"ChartMuseum (Anthropic No Tools)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"ChartMuseum (Anthropic Tools)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"ChartQAPro":{"url":"https://benchmarklist.com/benchmarks/chartqa_pro/","metric":"With tools score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ChartQAPro (Anthropic No Tools)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"ChartQAPro (Anthropic Tools)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Chartographer":{"url":"https://benchmarklist.com/benchmarks/chartographer_counterfactual_chart_generation_for_evaluating_vision_language_models/","metric":"ChartQA OA (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Chatbot Arena (Code)":{"url":"https://arena.ai/leaderboard/code","metric":"Elo","theoretical_min":null,"theoretical_max":null},"Chatbot Arena (Document)":{"url":"https://arena.ai/leaderboard/document","metric":"Elo","theoretical_min":null,"theoretical_max":null},"Chatbot Arena (Image Edit)":{"url":"https://arena.ai/leaderboard/image-edit","metric":"Elo","theoretical_min":null,"theoretical_max":null},"Chatbot Arena (Search)":{"url":"https://arena.ai/leaderboard/search","metric":"Arena Score","theoretical_min":null,"theoretical_max":null},"Chatbot Arena (Text)":{"url":"https://arena.ai/leaderboard/text","metric":"Elo","theoretical_min":null,"theoretical_max":null},"Chatbot Arena (Text-to-Image)":{"url":"https://arena.ai/leaderboard/text-to-image","metric":"Elo","theoretical_min":null,"theoretical_max":null},"Chatbot Arena (Vision)":{"url":"https://arena.ai/leaderboard/vision","metric":"Arena Score","theoretical_min":null,"theoretical_max":null},"CheXpercept":{"url":"https://benchmarklist.com/benchmarks/chexpercept/","metric":"Stage 1 (End-to-End) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Checkup2Action":{"url":"https://benchmarklist.com/benchmarks/checkup2action_a_multimodal_clinical_check_up_report_dataset_for_patient_oriented_action_card_generation/","metric":"Prob. rec. (problem recall) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ChemBench":{"url":"https://chembench.lamalab.org/","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Chess Bench LLM":{"url":"https://chessbenchllm.onrender.com/","metric":"Lichess Rating","theoretical_min":null,"theoretical_max":null},"Chess Puzzles (Epoch AI)":{"url":"https://epoch.ai/benchmarks/chess-puzzles","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"ChessImageBench":{"url":"https://github.com/JasperDekoninck/ChessImageBench","metric":"All Correct (%)","theoretical_min":0.0,"theoretical_max":100.0},"ChildAgentEval":{"url":"https://benchmarklist.com/benchmarks/childagenteval/","metric":"Total (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Chinese Classical Bench":{"url":"https://huggingface.co/spaces/gujilab/chinese-classical-bench-leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Chinese Classical Bench - Char-Gloss Judge":{"url":"https://huggingface.co/spaces/gujilab/chinese-classical-bench-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Chinese Classical Bench - Compress Efficiency":{"url":"https://huggingface.co/spaces/gujilab/chinese-classical-bench-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Chinese Classical Bench - Fill-In Exact":{"url":"https://huggingface.co/spaces/gujilab/chinese-classical-bench-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Chinese Classical Bench - Idiom-Source Book EM":{"url":"https://huggingface.co/spaces/gujilab/chinese-classical-bench-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Chinese Classical Bench - Punctuate Punct F1":{"url":"https://huggingface.co/spaces/gujilab/chinese-classical-bench-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Chinese Classical Bench - Translate Judge":{"url":"https://huggingface.co/spaces/gujilab/chinese-classical-bench-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"ChineseSafe Benchmark":{"url":"https://huggingface.co/spaces/SUSTech/ChineseSafe-Benchmark","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"CiteVQA":{"url":"https://benchmarklist.com/benchmarks/citevqa_benchmarking_evidence_attribution_for_trustworthy_document_intelligence/","metric":"Overall SAA (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Cited but Not Verified":{"url":"https://benchmarklist.com/benchmarks/cited_but_not_verified_parsing_and_evaluating_source_attribution_in_llm_deep_research_agents/","metric":"Relevant Content (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ClassEval":{"url":"https://fudanselab-classeval.github.io/","metric":"Class-Level Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"Claw-Anything":{"url":"https://benchmarklist.com/benchmarks/claw_anything_benchmarking_always_on_personal_assistants_with_broader_access_to_user_s_digital_world/","metric":"Pass@1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Claw-Eval-Live":{"url":"https://benchmarklist.com/benchmarks/claw_eval_live/","metric":"Pass Rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ClawArena":{"url":"https://www.clawarena.cc/leaderboard","metric":"Composite Reliability Score","theoretical_min":0.0,"theoretical_max":100.0},"ClawBench v1":{"url":"https://huggingface.co/datasets/TIGER-Lab/ClawBench","metric":"Reward Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"ClawBench v2":{"url":"https://huggingface.co/datasets/TIGER-Lab/ClawBench","metric":"Reward Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"ClawForge":{"url":"https://benchmarklist.com/benchmarks/clawforge/","metric":"Strict Acc. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ClawProBench":{"url":"https://benchmarklist.com/benchmarks/clawprobench/","metric":"Final Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Clembench Multimodal v1.6.5":{"url":"https://benchmarklist.com/benchmarks/clembench_multimodal_v1_6_5/","metric":"clemscore (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Clerk LLM Leaderboard":{"url":"https://clerk.com/llm-leaderboard","metric":"Avg score (%)","theoretical_min":0.0,"theoretical_max":100.0},"ClinPivot":{"url":"https://benchmarklist.com/benchmarks/clinpivot/","metric":"ClinPivot (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ClinicBench":{"url":"https://huggingface.co/spaces/fenglinliu/medical_llm_leaderboard","metric":"Average Score","theoretical_min":0.0,"theoretical_max":100.0},"ClockBench":{"url":"https://clockbench.ai/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"CocoaBench":{"url":"https://cocoabench.github.io/","metric":"Accuracy","theoretical_min":0.0,"theoretical_max":100.0},"CodeClash":{"url":"https://codeclash.ai","metric":"ELO","theoretical_min":null,"theoretical_max":null},"CodeClash - BattleSnake":{"url":"https://codeclash.ai","metric":"ELO","theoretical_min":null,"theoretical_max":null},"CodeClash - Core War":{"url":"https://codeclash.ai","metric":"ELO","theoretical_min":null,"theoretical_max":null},"CodeClash - Halite":{"url":"https://codeclash.ai","metric":"ELO","theoretical_min":null,"theoretical_max":null},"CodeClash - Poker":{"url":"https://codeclash.ai","metric":"ELO","theoretical_min":null,"theoretical_max":null},"CodeClash - Robocode":{"url":"https://codeclash.ai","metric":"ELO","theoretical_min":null,"theoretical_max":null},"CodeClash - Robot Rumble":{"url":"https://codeclash.ai","metric":"ELO","theoretical_min":null,"theoretical_max":null},"CodeClinic":{"url":"https://benchmarklist.com/benchmarks/codeclinic_evaluating_automation_of_coding_skills_for_clinical_reasoning_agents/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CodeElo":{"url":"https://codeelo-bench.github.io/","metric":"Elo Rating","theoretical_min":null,"theoretical_max":null},"CodeGolf Bench":{"url":"https://benchmarklist.com/benchmarks/codegolf_bench/","metric":"C++ Best %ile (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CodeScope":{"url":"https://haitianliu22.github.io/code-scope-benchmark/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"CodeScope - Code Generation":{"url":"https://haitianliu22.github.io/code-scope-benchmark/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"CodeScope - Code Repair":{"url":"https://haitianliu22.github.io/code-scope-benchmark/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"CodeScope - Code Summarization":{"url":"https://haitianliu22.github.io/code-scope-benchmark/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"CodeScope - Code Translation":{"url":"https://haitianliu22.github.io/code-scope-benchmark/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"CodeScope - Code Understanding":{"url":"https://haitianliu22.github.io/code-scope-benchmark/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"CodeScope - Program Synthesis":{"url":"https://haitianliu22.github.io/code-scope-benchmark/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"Codeforces ELO":{"url":"https://benchmarklist.com/benchmarks/codeforces_elo/","metric":"ELO (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CoffeeBench":{"url":"https://pub.sakana.ai/coffeebench/","metric":"Mean net income ($)","theoretical_min":null,"theoretical_max":null},"CombEval":{"url":"https://benchmarklist.com/benchmarks/combeval_a_framework_for_evaluating_combinatorial_counting_in_large_language_models/","metric":"Avg. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CommonWhy":{"url":"https://benchmarklist.com/benchmarks/commonwhy_a_dataset_for_evaluating_entity_based_causal_commonsense_reasoning_in_large_language_models/","metric":"BS-Prec (Long-Tail) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CommunityFact":{"url":"https://benchmarklist.com/benchmarks/communityfact_a_dynamic_multilingual_multi_domain_benchmark_for_misinformation_detection_in_the_wild/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Compl-AI Board":{"url":"https://huggingface.co/spaces/latticeflow/compl-ai-board","metric":"Average Compliance Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"ComplexFuncBench":{"url":"https://benchmarklist.com/benchmarks/complexfuncbench/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ComplexMCP":{"url":"https://benchmarklist.com/benchmarks/complexmcp/","metric":"Success Rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ConStory-Bench":{"url":"https://picrew.github.io/constory-bench.github.io/","metric":"CED errors per 10K words (lower is better)","theoretical_min":0.0,"theoretical_max":null},"Confabulation Leaderboard (Lechmazur)":{"url":"https://lechmazur.github.io/leaderboard1.html","metric":"Confabulation rate % (lower is better)","theoretical_min":0.0,"theoretical_max":100.0},"ConsumerSimBench":{"url":"https://benchmarklist.com/benchmarks/consumersimbench/","metric":"Avg (95% CI) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Context Arena MRCR (2-needle)":{"url":"https://contextarena.ai/","metric":"AUC@1M (%)","theoretical_min":0.0,"theoretical_max":100.0},"Context Arena MRCR (4-needle)":{"url":"https://contextarena.ai/","metric":"AUC@1M (%)","theoretical_min":0.0,"theoretical_max":100.0},"Context Arena MRCR (8-needle)":{"url":"https://contextarena.ai/","metric":"AUC@1M (%)","theoretical_min":0.0,"theoretical_max":100.0},"Context-Bench Filesystem":{"url":"https://leaderboard.letta.com/","metric":"Rubric Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Context-Bench Skills":{"url":"https://leaderboard.letta.com/","metric":"Task Completion (%)","theoretical_min":0.0,"theoretical_max":100.0},"ContextEcho":{"url":"https://benchmarklist.com/benchmarks/contextecho_a_benchmark_for_persona_drift_in_long_agentic_coding_sessions/","metric":"Delta (self-reported)","theoretical_min":null,"theoretical_max":null},"ContractBench":{"url":"https://benchmarklist.com/benchmarks/contractbench/","metric":"SR% (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ConvRe":{"url":"https://huggingface.co/spaces/3B-Group/ConvRe-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"ConvRe - Re2Text Easy":{"url":"https://huggingface.co/spaces/3B-Group/ConvRe-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"ConvRe - Re2Text Hard":{"url":"https://huggingface.co/spaces/3B-Group/ConvRe-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"ConvRe - Text2Re Easy":{"url":"https://huggingface.co/spaces/3B-Group/ConvRe-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"ConvRe - Text2Re Hard":{"url":"https://huggingface.co/spaces/3B-Group/ConvRe-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Cookie-Bench":{"url":"https://benchmarklist.com/benchmarks/cookie_bench_or_dataname/","metric":"React Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Correction Suppression Benchmark":{"url":"https://benchmarklist.com/benchmarks/correction_suppression_benchmark_300_false_premises/","metric":"Suppression Rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CounterCount":{"url":"https://benchmarklist.com/benchmarks/countercount_a_diagnostic_framework_for_counting_bias_in_vision_language_models/","metric":"Avg Acc (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Creation-MMBench":{"url":"https://huggingface.co/spaces/opencompass/openlmm_subjective_leaderboard","metric":"Overall Reward","theoretical_min":-100.0,"theoretical_max":100.0},"Creative Writing (Lechmazur)":{"url":"https://github.com/lechmazur/writing","metric":"Mean Score","theoretical_min":-10.0,"theoretical_max":10.0},"Creative Writing v3":{"url":"https://benchmarklist.com/benchmarks/creative_writing_v3/","metric":"Elo score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CritPt":{"url":"https://benchmarklist.com/benchmarks/critpt/","metric":"Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Crosby micro1 RedlineBench":{"url":"https://benchmarklist.com/benchmarks/redline_bench/","metric":"Turn-weighted score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CubeBench":{"url":"https://github.com/crabbixOCE/CubeBench","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"CulMind":{"url":"https://benchmarklist.com/benchmarks/culmind_benchmarking_multimodal_understanding_and_reasoning_in_chinese_cultural_heritage/","metric":"S (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CulturaQA":{"url":"https://benchmarklist.com/benchmarks/culturaqa/","metric":"CulturaQA (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CursorBench 3.1":{"url":"https://cursor.com/evals","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"CyBench":{"url":"https://cybench.github.io/","metric":"End-to-End % Solved","theoretical_min":0.0,"theoretical_max":100.0},"CyScenarioBench":{"url":"https://epochai.substack.com/p/are-mythos-cyber-capabilities-overhyped","metric":"Average Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"CyberBench (NLP)":{"url":"https://huggingface.co/spaces/FrontierAICybersecurity/Cybersecurity_leaderboard","metric":"Avg Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"CyberGym":{"url":"https://benchmarklist.com/benchmarks/cybergym/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CyberMetric":{"url":"https://huggingface.co/spaces/FrontierAICybersecurity/Cybersecurity_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"CyberSecEval":{"url":"https://benchmarklist.com/benchmarks/cyberseceval/","metric":"Average Injection Success Rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"CyberSecEval-3":{"url":"https://huggingface.co/spaces/FrontierAICybersecurity/Cybersecurity_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"CyberSecEval2 Interpreter Abuse":{"url":"https://github.com/UKGovernmentBEIS/inspect_evals/tree/main/src/inspect_evals/cyberseceval_2/interpreter_abuse","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"CyberSecEval2 Prompt Injection":{"url":"https://github.com/UKGovernmentBEIS/inspect_evals/tree/main/src/inspect_evals/cyberseceval_2/prompt_injection","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"CyberSecEval2 Vulnerability Exploit":{"url":"https://github.com/UKGovernmentBEIS/inspect_evals/tree/main/src/inspect_evals/cyberseceval_2/vulnerability_exploit","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"DABstep":{"url":"https://huggingface.co/spaces/adyen/DABstep","metric":"Hard Level Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"DEEPSYNTH":{"url":"https://benchmarklist.com/benchmarks/deepsynth/","metric":"F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"DITING":{"url":"https://huggingface.co/spaces/NextGenWhu/DITING-leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":null},"DITING - Cultural Safety":{"url":"https://huggingface.co/spaces/NextGenWhu/DITING-leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":null},"DITING - Idiom Translation":{"url":"https://huggingface.co/spaces/NextGenWhu/DITING-leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":null},"DITING - Lexical Ambiguity":{"url":"https://huggingface.co/spaces/NextGenWhu/DITING-leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":null},"DITING - Tense Consistency":{"url":"https://huggingface.co/spaces/NextGenWhu/DITING-leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":null},"DITING - Terminology Localization":{"url":"https://huggingface.co/spaces/NextGenWhu/DITING-leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":null},"DITING - Zero Pronoun Translation":{"url":"https://huggingface.co/spaces/NextGenWhu/DITING-leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":null},"DRACO":{"url":"https://leaderboard.steel.dev/leaderboards/draco/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"DRInQ":{"url":"https://benchmarklist.com/benchmarks/drinq_evaluating_conversational_implicature_with_controlled_context_variation/","metric":"Vanilla (self-reported)","theoretical_min":0.0,"theoretical_max":null},"DROP":{"url":"https://github.com/openai/simple-evals","metric":"F1 Score","theoretical_min":0.0,"theoretical_max":100.0},"DVMap":{"url":"https://benchmarklist.com/benchmarks/dvmap_fine_grained_pluralistic_value_alignment_via_high_consensus_demographic_value_mapping/","metric":"Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Darija Chatbot Arena":{"url":"https://huggingface.co/spaces/atlasia/darija-chatbot-arena","metric":"Elo Rating","theoretical_min":null,"theoretical_max":null},"DataClawBench":{"url":"https://benchmarklist.com/benchmarks/dataclawbench_an_agent_benchmark_for_exploratory_real_world_financial_data_analysis/","metric":"Overall Acc. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Deception Effectiveness (Lechmazur)":{"url":"https://github.com/lechmazur/deception","metric":"Deception Score","theoretical_min":0.0,"theoretical_max":null},"Deception Resistance (Lechmazur)":{"url":"https://github.com/lechmazur/deception","metric":"Vulnerability Score (lower is better)","theoretical_min":0.0,"theoretical_max":null},"Decomposing and Measuring Evaluation Awareness":{"url":"https://benchmarklist.com/benchmarks/decomposing_and_measuring_evaluation_awareness/","metric":"HarmBench Saf. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"DeepResearch Bench":{"url":"https://huggingface.co/spaces/muset-ai/DeepResearch-Bench-Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"DeepResearch Bench - Citation Accuracy":{"url":"https://huggingface.co/spaces/muset-ai/DeepResearch-Bench-Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"DeepResearch Bench - Comprehensiveness":{"url":"https://huggingface.co/spaces/muset-ai/DeepResearch-Bench-Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"DeepResearch Bench - Effective Citations":{"url":"https://huggingface.co/spaces/muset-ai/DeepResearch-Bench-Leaderboard","metric":"Effective Citations","theoretical_min":0.0,"theoretical_max":null},"DeepResearch Bench - Insight":{"url":"https://huggingface.co/spaces/muset-ai/DeepResearch-Bench-Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"DeepResearch Bench - Instruction Following":{"url":"https://huggingface.co/spaces/muset-ai/DeepResearch-Bench-Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"DeepResearch Bench - Readability":{"url":"https://huggingface.co/spaces/muset-ai/DeepResearch-Bench-Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"DeepResearchBench":{"url":"https://deepresearch-bench.github.io/","metric":"Average Score","theoretical_min":0.0,"theoretical_max":100.0},"DeepSWE":{"url":"https://deepswe.datacurve.ai/","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"DeepSearchQA":{"url":"https://benchmarklist.com/benchmarks/deepsearchqa/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"DeepWeb-Bench":{"url":"https://benchmarklist.com/benchmarks/deepweb_bench_a_deep_research_benchmark_demanding_massive_cross_source_evidence_and_long_horizon_derivation/","metric":"Overall score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Defects4J":{"url":"https://benchmarklist.com/benchmarks/defects4j/","metric":"Defects4J Plausible @1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"DentalGPT Dental Eval Suite":{"url":"https://benchmarklist.com/benchmarks/dentalgpt_dental_eval_suite/","metric":"Average Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Design Arena (3D)":{"url":"https://www.designarena.ai/","metric":"Elo","theoretical_min":null,"theoretical_max":null},"Design Arena (Data Viz)":{"url":"https://www.designarena.ai/","metric":"Elo","theoretical_min":null,"theoretical_max":null},"Design Arena (Game Dev)":{"url":"https://www.designarena.ai/","metric":"Elo","theoretical_min":null,"theoretical_max":null},"Design Arena (SVG)":{"url":"https://www.designarena.ai/","metric":"Elo","theoretical_min":null,"theoretical_max":null},"Design Arena (UI Components)":{"url":"https://www.designarena.ai/","metric":"Elo","theoretical_min":null,"theoretical_max":null},"Design Arena (Website)":{"url":"https://www.designarena.ai/","metric":"Elo","theoretical_min":null,"theoretical_max":null},"DiBiMT":{"url":"https://nlp.uniroma1.it/dibimt/","metric":"Avg Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"DiagFlowBench":{"url":"https://benchmarklist.com/benchmarks/diagflowbench_evaluating_how_language_models_handle_off_procedure_inputs_in_grounded_diagnostic_dialogue/","metric":"Step Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Diagnosing LLM Arbitration Behavior over Pre-e":{"url":"https://benchmarklist.com/benchmarks/diagnosing_llm_arbitration_behavior_over_pre_evidence_epistemic_states_in_rag_based_fact_checking/","metric":"Margin (self-reported)","theoretical_min":null,"theoretical_max":null},"DiagnosticIQ":{"url":"https://benchmarklist.com/benchmarks/diagnosticiq_a_benchmark_for_llm_based_industrial_maintenance_action_recommendation_from_symbolic_rules/","metric":"Macro % D.IQ (self-reported)","theoretical_min":0.0,"theoretical_max":null},"DiffCap-Bench":{"url":"https://benchmarklist.com/benchmarks/diffcap_bench_a_comprehensive_challenging_robust_benchmark_for_image_difference_captioning/","metric":"F1* (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Diplomacy: Betrayal Tendency":{"url":"https://goodstartlabs.com/leaderboards/diplomacy","metric":"Betrayal Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Diplomacy: Overall Performance":{"url":"https://goodstartlabs.com/leaderboards/diplomacy","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"Diplomacy: Steerability":{"url":"https://goodstartlabs.com/leaderboards/diplomacy","metric":"Steerability Score","theoretical_min":null,"theoretical_max":null},"DisasterBench":{"url":"https://benchmarklist.com/benchmarks/disasterbench_benchmarking_llm_planning_under_typed_tool_interface_constraints/","metric":"Exact-match Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"DiscoX":{"url":"https://randomtutu.github.io/DiscoX/","metric":"Overall Score","theoretical_min":0.0,"theoretical_max":100.0},"DiscoverPhysics":{"url":"https://benchmarklist.com/benchmarks/discoverphysics/","metric":"Mean Explanation Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Disentangling Language Roles in Multilingual L":{"url":"https://benchmarklist.com/benchmarks/disentangling_language_roles_in_multilingual_llm_task_execution/","metric":"JointSuccess (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Divergent Thinking":{"url":"https://github.com/lechmazur/divergent","metric":"Divergence Score","theoretical_min":0.0,"theoretical_max":10.0},"Do Agents Know What They Can't Do? Evaluating":{"url":"https://benchmarklist.com/benchmarks/do_agents_know_what_they_can_t_do_evaluating_feasibility_awareness_in_tool_using_agents/","metric":"Avg. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Do Coding Agents Understand Least-Privilege Au":{"url":"https://benchmarklist.com/benchmarks/do_coding_agents_understand_least_privilege_authorization/","metric":"TSR (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Do Enterprise Systems Need Learned World Model":{"url":"https://benchmarklist.com/benchmarks/do_enterprise_systems_need_learned_world_models_the_importance_of_context_to_infer_dynamics/","metric":"CascadeBench IoU w/ BR (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Do LLMs Build World Models From Text? A Multil":{"url":"https://benchmarklist.com/benchmarks/do_llms_build_world_models_from_text_a_multilingual_diagnostic_of_spatial_reasoning/","metric":"L3 (viewpoint) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"DocScope":{"url":"https://benchmarklist.com/benchmarks/docscope/","metric":"ACC All (self-reported)","theoretical_min":0.0,"theoretical_max":null},"DomainCodeBench":{"url":"https://huggingface.co/datasets/umaimakhan01/domain-code-bench","metric":"Composite Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Done, But Not Sure":{"url":"https://benchmarklist.com/benchmarks/done_but_not_sure_disentangling_world_completion_from_self_termination_in_embodied_agents/","metric":"B All (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Dr. DocBench":{"url":"https://benchmarklist.com/benchmarks/dr_docbench_a_comprehensive_benchmark_for_expert_level_and_difficult_document_parsing/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"DramaBench":{"url":"https://dramabench.pages.dev/","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"DrugDiscoveryBench":{"url":"https://scale.com/blog/drugdiscoverybench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"DuckDB-NSQL":{"url":"https://huggingface.co/spaces/duckdb-nsql-hub/DuckDB-NSQL-Leaderboard","metric":"Execution Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ducky Bench (Africa)":{"url":"https://ducky-bench.joinity.site/leaderboard.php?test=3","metric":"ELO","theoretical_min":null,"theoretical_max":null},"Ducky Bench (Saxo Frog)":{"url":"https://ducky-bench.joinity.site/leaderboard.php?test=2","metric":"ELO","theoretical_min":null,"theoretical_max":null},"Ducky Bench (Stabby Quack)":{"url":"https://ducky-bench.joinity.site/leaderboard.php?test=1","metric":"ELO","theoretical_min":null,"theoretical_max":null},"DuelLab Overall":{"url":"https://benchmarklist.com/benchmarks/duellab_overall/","metric":"Avg score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"DystopiaBench":{"url":"https://dystopiabench.com/","metric":"Dystopian Compliance Score (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"ECLeKTic":{"url":"https://benchmarklist.com/benchmarks/eclektic/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"EComAgentBench":{"url":"https://benchmarklist.com/benchmarks/ecomagentbench_benchmarking_shopping_agents_on_long_horizon_tasks_with_distributed_hidden_intent/","metric":"Acc. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"EHR-Complex":{"url":"https://benchmarklist.com/benchmarks/ehr_complex/","metric":"Avg. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"EHRBench":{"url":"https://benchmarklist.com/benchmarks/ehrbench_an_automated_and_reliable_ehr_based_benchmark_for_clinical_decision_making_with_llms/","metric":"Overall Acc (%) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"EIBench":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"Recall","theoretical_min":0.0,"theoretical_max":100.0},"EIBench (complex)":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"Recall","theoretical_min":0.0,"theoretical_max":100.0},"ENAMEL":{"url":"https://benchmarklist.com/benchmarks/enamel/","metric":"eff@1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"EO-Gym":{"url":"https://benchmarklist.com/benchmarks/eo_gym_a_multimodal_interactive_environment_for_earth_observation_agents/","metric":"Pass@3 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"EQ-Bench":{"url":"https://benchmarklist.com/benchmarks/eq_bench/","metric":"Normalized Elo (self-reported)","theoretical_min":0.0,"theoretical_max":null},"EQ-Bench 3":{"url":"https://eqbench.com/","metric":"Elo","theoretical_min":null,"theoretical_max":null},"EQ-Bench Creative Writing v3":{"url":"https://eqbench.com/creative_writing.html","metric":"Elo","theoretical_min":null,"theoretical_max":null},"EQ-Bench Longform Writing":{"url":"https://eqbench.com/creative_writing_longform.html","metric":"Writing Score (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"EQUITRIAGE":{"url":"https://benchmarklist.com/benchmarks/equitriage_a_fairness_audit_of_gender_bias_in_llm_based_emergency_department_triage/","metric":"Exact% (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ERGeoBench":{"url":"https://benchmarklist.com/benchmarks/ergeobench_a_comprehensive_benchmark_for_embodied_reasoning_and_geo_localization_in_multimodal_large_language_models/","metric":"Street (%) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"EVALITA - MAIA-GEN":{"url":"https://huggingface.co/datasets/evalitahf/evalita_llm_results","metric":"CPS","theoretical_min":0.0,"theoretical_max":100.0},"EVALITA - MAIA-MC":{"url":"https://huggingface.co/datasets/evalitahf/evalita_llm_results","metric":"CPS","theoretical_min":0.0,"theoretical_max":100.0},"EVALITA - admission-test":{"url":"https://huggingface.co/datasets/evalitahf/evalita_llm_results","metric":"CPS","theoretical_min":0.0,"theoretical_max":100.0},"EVALITA - evalita NER":{"url":"https://huggingface.co/datasets/evalitahf/evalita_llm_results","metric":"CPS","theoretical_min":0.0,"theoretical_max":100.0},"EVALITA - faq":{"url":"https://huggingface.co/datasets/evalitahf/evalita_llm_results","metric":"CPS","theoretical_min":0.0,"theoretical_max":100.0},"EVALITA - hate-speech-detection":{"url":"https://huggingface.co/datasets/evalitahf/evalita_llm_results","metric":"CPS","theoretical_min":0.0,"theoretical_max":100.0},"EVALITA - lexical-substitution":{"url":"https://huggingface.co/datasets/evalitahf/evalita_llm_results","metric":"CPS","theoretical_min":0.0,"theoretical_max":100.0},"EVALITA - relation-extraction":{"url":"https://huggingface.co/datasets/evalitahf/evalita_llm_results","metric":"CPS","theoretical_min":0.0,"theoretical_max":100.0},"EVALITA - sentiment-analysis":{"url":"https://huggingface.co/datasets/evalitahf/evalita_llm_results","metric":"CPS","theoretical_min":0.0,"theoretical_max":100.0},"EVALITA - summarization-fanpage":{"url":"https://huggingface.co/datasets/evalitahf/evalita_llm_results","metric":"CPS","theoretical_min":0.0,"theoretical_max":100.0},"EVALITA - text-entailment":{"url":"https://huggingface.co/datasets/evalitahf/evalita_llm_results","metric":"CPS","theoretical_min":0.0,"theoretical_max":100.0},"EVALITA - word-in-context":{"url":"https://huggingface.co/datasets/evalitahf/evalita_llm_results","metric":"CPS","theoretical_min":0.0,"theoretical_max":100.0},"EVALITA LLM Leaderboard":{"url":"https://huggingface.co/datasets/evalitahf/evalita_llm_results","metric":"Average CPS","theoretical_min":0.0,"theoretical_max":100.0},"EVMbench":{"url":"https://github.com/paradigmxyz/evmbench","metric":"Avg Score (Detect+Patch+Exploit) (%)","theoretical_min":0.0,"theoretical_max":100.0},"EconLogicQA":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"Accuracy","theoretical_min":0.0,"theoretical_max":1.0},"EdgeBench":{"url":"https://edge-bench.org/","metric":"Score @12h (134 tasks)","theoretical_min":null,"theoretical_max":null},"EduGuardBench":{"url":"https://benchmarklist.com/benchmarks/eduguardbench/","metric":"RFS (self-reported)","theoretical_min":0.0,"theoretical_max":null},"EffiBench - NET":{"url":"https://huggingface.co/spaces/EffiBench/effibench-leaderboard","metric":"Normalized Execution Time","theoretical_min":0.0,"theoretical_max":null},"EffiBench - NMU":{"url":"https://huggingface.co/spaces/EffiBench/effibench-leaderboard","metric":"Normalized Memory Usage","theoretical_min":0.0,"theoretical_max":null},"Ego2World":{"url":"https://benchmarklist.com/benchmarks/ego2world/","metric":"Goal Tasks (self-reported)","theoretical_min":0.0,"theoretical_max":null},"EgoCoT-Bench":{"url":"https://benchmarklist.com/benchmarks/egocot_bench_benchmarking_grounded_and_verifiable_operation_centric_chain_of_thought_reasoning_for_mllms/","metric":"Mean (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Eleusis Benchmark":{"url":"https://huggingface.co/spaces/huggingface/eleusis-benchmark","metric":"Avg Floored Score","theoretical_min":0.0,"theoretical_max":null},"Elimination Game (Lechmazur)":{"url":"https://github.com/lechmazur/elimination_game","metric":"TrueSkill μ","theoretical_min":null,"theoretical_max":null},"EmbodiedBench ALFRED":{"url":"https://embodiedbench.github.io/","metric":"Avg Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EmbodiedBench Habitat":{"url":"https://embodiedbench.github.io/","metric":"Avg Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EmoBench-M":{"url":"https://github.com/Emo-gml/EmoBench-M","metric":"Average Score","theoretical_min":0.0,"theoretical_max":100.0},"Emotional intelligence in large language model":{"url":"https://benchmarklist.com/benchmarks/emotional_intelligence_in_large_language_models_is_fragmented_across_perception_cognition_and_interaction/","metric":"Total Kappa (self-reported)","theoretical_min":0.0,"theoretical_max":null},"EmpathyBench":{"url":"https://www.empathybench.com/","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EnactToM":{"url":"https://benchmarklist.com/benchmarks/enacttom_an_evolving_benchmark_for_functional_theory_of_mind_in_embodied_agents/","metric":"Avg Standard Functional (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Enhancing Agent Safety Judgment":{"url":"https://benchmarklist.com/benchmarks/enhancing_agent_safety_judgment_controlled_benchmark_rewriting_and_analogical_reasoning_for_deceptive_out_of_distribution_scenarios/","metric":"Original (Seed) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"EnigmaEval":{"url":"https://benchmarklist.com/benchmarks/enigma_eval/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"EnterpriseArena":{"url":"https://benchmarklist.com/benchmarks/enterprisearena/","metric":"Full Survival % (self-reported)","theoretical_min":0.0,"theoretical_max":null},"EnterpriseClawBench (Claude Code)":{"url":"https://frontisai.github.io/EnterpriseClawBench/","metric":"Primary score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EnterpriseClawBench (Codex)":{"url":"https://frontisai.github.io/EnterpriseClawBench/","metric":"Primary score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EnterpriseClawBench (DeepAgents)":{"url":"https://frontisai.github.io/EnterpriseClawBench/","metric":"Primary score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EnterpriseClawBench (Hermes)":{"url":"https://frontisai.github.io/EnterpriseClawBench/","metric":"Primary score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EnterpriseClawBench (OpenClaw)":{"url":"https://frontisai.github.io/EnterpriseClawBench/","metric":"Primary score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EnterpriseOps-Gym":{"url":"https://benchmarklist.com/benchmarks/enterpriseops_gym/","metric":"Task Success Rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"EnvSimBench":{"url":"https://benchmarklist.com/benchmarks/envsimbench_a_benchmark_for_evaluating_and_improving_llm_based_environment_simulation/","metric":"Cm Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Episodic Memory":{"url":"https://github.com/ahstat/episodic-memory-benchmark","metric":"Simple Recall (%)","theoretical_min":0.0,"theoretical_max":100.0},"Epoch AI - Adversarial Nli":{"url":"https://epoch.ai/benchmarks/adversarial-nli","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"Epoch AI - Algotune":{"url":"https://epoch.ai/benchmarks/algotune","metric":"Score","theoretical_min":0.0,"theoretical_max":null},"Epoch AI - Apex Agents":{"url":"https://epoch.ai/benchmarks/apex-agents","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"Epoch AI - Blueprint Bench 2":{"url":"https://epoch.ai/benchmarks/blueprint-bench-2","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"Epoch AI - Common Sense Qa 2":{"url":"https://epoch.ai/benchmarks/common-sense-qa-2","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"Epoch AI - Critpt":{"url":"https://epoch.ai/benchmarks/critpt","metric":"Score","theoretical_min":0.0,"theoretical_max":null},"Epoch AI - Cursorbench":{"url":"https://epoch.ai/benchmarks/cursorbench","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"Epoch AI - ECI":{"url":"https://epoch.ai/benchmarks/eci","metric":"ECI Score","theoretical_min":null,"theoretical_max":null},"Epoch AI - Lambada":{"url":"https://epoch.ai/benchmarks/lambada","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"Epoch AI - Rli":{"url":"https://epoch.ai/benchmarks/rli","metric":"Score","theoretical_min":0.0,"theoretical_max":null},"Epoch AI - Scicode":{"url":"https://epoch.ai/benchmarks/scicode","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"Epoch AI - Science Qa":{"url":"https://epoch.ai/benchmarks/science-qa","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"Epoch AI - Superglue":{"url":"https://epoch.ai/benchmarks/superglue","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"Epoch AI - Vending Bench 2":{"url":"https://epoch.ai/benchmarks/vending-bench-2","metric":"Score","theoretical_min":null,"theoretical_max":null},"ErdosBench":{"url":"https://www.ulam.ai/erdosbench","metric":"Judge Average (0-4)","theoretical_min":null,"theoretical_max":null},"EsoBench":{"url":"https://caseys-evals.com/esobench","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"EsoLang-Bench":{"url":"https://esolang-bench.vercel.app/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Albanian":{"url":"https://euroeval.com/leaderboards/albanian/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Albanian Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/albanian/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Albanian Knowledge":{"url":"https://euroeval.com/leaderboards/albanian/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Albanian NLU":{"url":"https://euroeval.com/leaderboards/albanian/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Albanian NLU - MMS SQ":{"url":"https://euroeval.com/leaderboards/albanian/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Albanian NLU - MultiWikiQA SQ":{"url":"https://euroeval.com/leaderboards/albanian/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Albanian NLU - ScaLA SQ":{"url":"https://euroeval.com/leaderboards/albanian/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Albanian NLU - WikiANN SQ":{"url":"https://euroeval.com/leaderboards/albanian/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Albanian Summarization - LR SUM SQ":{"url":"https://euroeval.com/leaderboards/albanian/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Bosnian":{"url":"https://euroeval.com/leaderboards/bosnian/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Bosnian NLU":{"url":"https://euroeval.com/leaderboards/bosnian/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Bosnian NLU - MMS BS":{"url":"https://euroeval.com/leaderboards/bosnian/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Bosnian NLU - Multi Wiki QA BS":{"url":"https://euroeval.com/leaderboards/bosnian/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Bosnian NLU - WikiANN BS":{"url":"https://euroeval.com/leaderboards/bosnian/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Bosnian Summarization - LR SUM BS":{"url":"https://euroeval.com/leaderboards/bosnian/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Bulgarian":{"url":"https://euroeval.com/leaderboards/bulgarian/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Bulgarian Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/bulgarian/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Bulgarian Knowledge":{"url":"https://euroeval.com/leaderboards/bulgarian/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Bulgarian NLU":{"url":"https://euroeval.com/leaderboards/bulgarian/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Bulgarian NLU - BG NER Bsnlp":{"url":"https://euroeval.com/leaderboards/bulgarian/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Bulgarian NLU - Cinexio":{"url":"https://euroeval.com/leaderboards/bulgarian/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Bulgarian NLU - MultiWikiQA BG":{"url":"https://euroeval.com/leaderboards/bulgarian/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Bulgarian NLU - ScaLA BG":{"url":"https://euroeval.com/leaderboards/bulgarian/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Catalan":{"url":"https://euroeval.com/leaderboards/catalan/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Catalan Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/catalan/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Catalan Knowledge":{"url":"https://euroeval.com/leaderboards/catalan/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Catalan NLU":{"url":"https://euroeval.com/leaderboards/catalan/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Catalan NLU - Guia CAT":{"url":"https://euroeval.com/leaderboards/catalan/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Catalan NLU - MultiWikiQA CA":{"url":"https://euroeval.com/leaderboards/catalan/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Catalan NLU - ScaLA CA":{"url":"https://euroeval.com/leaderboards/catalan/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Catalan NLU - WikiANN CA":{"url":"https://euroeval.com/leaderboards/catalan/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Catalan Summarization - Dacsa CA":{"url":"https://euroeval.com/leaderboards/catalan/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Croatian":{"url":"https://euroeval.com/leaderboards/croatian/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Croatian Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/croatian/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Croatian Knowledge":{"url":"https://euroeval.com/leaderboards/croatian/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Croatian NLU":{"url":"https://euroeval.com/leaderboards/croatian/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Croatian NLU - MMS HR":{"url":"https://euroeval.com/leaderboards/croatian/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Croatian NLU - Multi Wiki QA HR":{"url":"https://euroeval.com/leaderboards/croatian/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Croatian NLU - ScaLA HR":{"url":"https://euroeval.com/leaderboards/croatian/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Croatian NLU - WikiANN HR":{"url":"https://euroeval.com/leaderboards/croatian/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Czech":{"url":"https://euroeval.com/leaderboards/czech/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Czech Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/czech/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Czech Knowledge":{"url":"https://euroeval.com/leaderboards/czech/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Czech NLU":{"url":"https://euroeval.com/leaderboards/czech/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Czech NLU - CS GEC":{"url":"https://euroeval.com/leaderboards/czech/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Czech NLU - CSFD Sentiment":{"url":"https://euroeval.com/leaderboards/czech/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Czech NLU - PONER":{"url":"https://euroeval.com/leaderboards/czech/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Czech NLU - SQAD":{"url":"https://euroeval.com/leaderboards/czech/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Czech Summarization - Czech News":{"url":"https://euroeval.com/leaderboards/czech/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Danish":{"url":"https://euroeval.com/leaderboards/danish/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Danish Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/danish/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Danish Knowledge":{"url":"https://euroeval.com/leaderboards/danish/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Danish NLU":{"url":"https://euroeval.com/leaderboards/danish/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Danish NLU - Angry Tweets":{"url":"https://euroeval.com/leaderboards/danish/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Danish NLU - Dansk":{"url":"https://euroeval.com/leaderboards/danish/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Danish NLU - Multi Wiki QA DA":{"url":"https://euroeval.com/leaderboards/danish/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Danish NLU - ScaLA DA":{"url":"https://euroeval.com/leaderboards/danish/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Danish Summarization - Nordjylland News":{"url":"https://euroeval.com/leaderboards/danish/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Dutch":{"url":"https://euroeval.com/leaderboards/dutch/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Dutch Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/dutch/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Dutch Knowledge":{"url":"https://euroeval.com/leaderboards/dutch/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Dutch NLU":{"url":"https://euroeval.com/leaderboards/dutch/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Dutch NLU - CoNLL NL":{"url":"https://euroeval.com/leaderboards/dutch/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Dutch NLU - DBRD":{"url":"https://euroeval.com/leaderboards/dutch/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Dutch NLU - SQuAD NL":{"url":"https://euroeval.com/leaderboards/dutch/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Dutch NLU - ScaLA NL":{"url":"https://euroeval.com/leaderboards/dutch/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Dutch Simplification - Duidelijke Taal":{"url":"https://euroeval.com/leaderboards/dutch/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Dutch Summarization - Wiki Lingua NL":{"url":"https://euroeval.com/leaderboards/dutch/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval English":{"url":"https://euroeval.com/leaderboards/english/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval English Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/english/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval English Knowledge":{"url":"https://euroeval.com/leaderboards/english/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval English NLU":{"url":"https://euroeval.com/leaderboards/english/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval English NLU - CoNLL EN":{"url":"https://euroeval.com/leaderboards/english/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval English NLU - SQuAD":{"url":"https://euroeval.com/leaderboards/english/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval English NLU - SST-5":{"url":"https://euroeval.com/leaderboards/english/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval English NLU - ScaLA EN":{"url":"https://euroeval.com/leaderboards/english/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval English Summarization - CNN Dailymail":{"url":"https://euroeval.com/leaderboards/english/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Estonian":{"url":"https://euroeval.com/leaderboards/estonian/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Estonian Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/estonian/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Estonian Knowledge":{"url":"https://euroeval.com/leaderboards/estonian/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Estonian NLU":{"url":"https://euroeval.com/leaderboards/estonian/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Estonian NLU - EstNER":{"url":"https://euroeval.com/leaderboards/estonian/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Estonian NLU - Estonian Valence":{"url":"https://euroeval.com/leaderboards/estonian/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Estonian NLU - Grammar ET":{"url":"https://euroeval.com/leaderboards/estonian/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Estonian NLU - MultiWikiQA ET":{"url":"https://euroeval.com/leaderboards/estonian/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Estonian Summarization - ERR News":{"url":"https://euroeval.com/leaderboards/estonian/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Faroese":{"url":"https://euroeval.com/leaderboards/faroese/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Faroese NLU - FONE":{"url":"https://euroeval.com/leaderboards/faroese/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Faroese NLU - FoQA":{"url":"https://euroeval.com/leaderboards/faroese/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Faroese NLU - FoSent":{"url":"https://euroeval.com/leaderboards/faroese/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Faroese NLU - ScaLA FO":{"url":"https://euroeval.com/leaderboards/faroese/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Finnish":{"url":"https://euroeval.com/leaderboards/finnish/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Finnish Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/finnish/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Finnish NLU":{"url":"https://euroeval.com/leaderboards/finnish/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Finnish NLU - ScaLA FI":{"url":"https://euroeval.com/leaderboards/finnish/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Finnish NLU - Scandisent FI":{"url":"https://euroeval.com/leaderboards/finnish/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Finnish NLU - Turku NER FI":{"url":"https://euroeval.com/leaderboards/finnish/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Finnish NLU - Tydiqa FI":{"url":"https://euroeval.com/leaderboards/finnish/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Finnish Summarization - Xlsum FI":{"url":"https://euroeval.com/leaderboards/finnish/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval French":{"url":"https://euroeval.com/leaderboards/french/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval French Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/french/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval French Knowledge":{"url":"https://euroeval.com/leaderboards/french/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval French NLU":{"url":"https://euroeval.com/leaderboards/french/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval French NLU - Allocine":{"url":"https://euroeval.com/leaderboards/french/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval French NLU - Eltec":{"url":"https://euroeval.com/leaderboards/french/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval French NLU - FQuAD":{"url":"https://euroeval.com/leaderboards/french/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval French NLU - ScaLA FR":{"url":"https://euroeval.com/leaderboards/french/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval French Summarization - Orange SUM":{"url":"https://euroeval.com/leaderboards/french/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval German":{"url":"https://euroeval.com/leaderboards/german/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval German Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/german/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval German Knowledge":{"url":"https://euroeval.com/leaderboards/german/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval German NLU":{"url":"https://euroeval.com/leaderboards/german/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval German NLU - GermEval":{"url":"https://euroeval.com/leaderboards/german/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval German NLU - Germanquad":{"url":"https://euroeval.com/leaderboards/german/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval German NLU - Sb10K":{"url":"https://euroeval.com/leaderboards/german/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval German NLU - ScaLA DE":{"url":"https://euroeval.com/leaderboards/german/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval German Summarization - Mlsum DE":{"url":"https://euroeval.com/leaderboards/german/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Greek":{"url":"https://euroeval.com/leaderboards/greek/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Greek Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/greek/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Greek Knowledge":{"url":"https://euroeval.com/leaderboards/greek/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Greek NLU":{"url":"https://euroeval.com/leaderboards/greek/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Greek NLU - ELNER":{"url":"https://euroeval.com/leaderboards/greek/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Greek NLU - Greek SA":{"url":"https://euroeval.com/leaderboards/greek/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Greek NLU - MultiWikiQA EL":{"url":"https://euroeval.com/leaderboards/greek/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Greek NLU - ScaLA EL":{"url":"https://euroeval.com/leaderboards/greek/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Greek Summarization - Greek Wikipedia":{"url":"https://euroeval.com/leaderboards/greek/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Hungarian":{"url":"https://euroeval.com/leaderboards/hungarian/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Hungarian Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/hungarian/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Hungarian Knowledge":{"url":"https://euroeval.com/leaderboards/hungarian/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Hungarian NLU":{"url":"https://euroeval.com/leaderboards/hungarian/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Hungarian NLU - HuSST":{"url":"https://euroeval.com/leaderboards/hungarian/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Hungarian NLU - MultiWikiQA HU":{"url":"https://euroeval.com/leaderboards/hungarian/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Hungarian NLU - ScaLA HU":{"url":"https://euroeval.com/leaderboards/hungarian/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Hungarian NLU - Szeged NER":{"url":"https://euroeval.com/leaderboards/hungarian/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Hungarian Summarization - Hunsum":{"url":"https://euroeval.com/leaderboards/hungarian/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Icelandic":{"url":"https://euroeval.com/leaderboards/icelandic/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Icelandic Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/icelandic/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Icelandic Knowledge":{"url":"https://euroeval.com/leaderboards/icelandic/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Icelandic NLU":{"url":"https://euroeval.com/leaderboards/icelandic/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Icelandic NLU - Hotter and Colder Sentiment":{"url":"https://euroeval.com/leaderboards/icelandic/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Icelandic NLU - MIM-GOLD NER":{"url":"https://euroeval.com/leaderboards/icelandic/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Icelandic NLU - NQII":{"url":"https://euroeval.com/leaderboards/icelandic/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Icelandic NLU - ScaLA IS":{"url":"https://euroeval.com/leaderboards/icelandic/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Icelandic Summarization - RRN":{"url":"https://euroeval.com/leaderboards/icelandic/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Italian":{"url":"https://euroeval.com/leaderboards/italian/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Italian Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/italian/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Italian Knowledge":{"url":"https://euroeval.com/leaderboards/italian/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Italian NLU":{"url":"https://euroeval.com/leaderboards/italian/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Italian NLU - MultiNERD IT":{"url":"https://euroeval.com/leaderboards/italian/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Italian NLU - SQuAD IT":{"url":"https://euroeval.com/leaderboards/italian/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Italian NLU - ScaLA IT":{"url":"https://euroeval.com/leaderboards/italian/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Italian NLU - Sentipolc16":{"url":"https://euroeval.com/leaderboards/italian/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Italian Summarization - Ilpost SUM":{"url":"https://euroeval.com/leaderboards/italian/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Latvian":{"url":"https://euroeval.com/leaderboards/latvian/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Latvian Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/latvian/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Latvian Knowledge":{"url":"https://euroeval.com/leaderboards/latvian/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Latvian NLU":{"url":"https://euroeval.com/leaderboards/latvian/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Latvian NLU - Fullstack NER LV":{"url":"https://euroeval.com/leaderboards/latvian/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Latvian NLU - Latvian Twitter Sentiment":{"url":"https://euroeval.com/leaderboards/latvian/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Latvian NLU - MultiWikiQA LV":{"url":"https://euroeval.com/leaderboards/latvian/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Latvian NLU - ScaLA LV":{"url":"https://euroeval.com/leaderboards/latvian/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Lithuanian":{"url":"https://euroeval.com/leaderboards/lithuanian/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Lithuanian Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/lithuanian/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Lithuanian Knowledge":{"url":"https://euroeval.com/leaderboards/lithuanian/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Lithuanian NLU":{"url":"https://euroeval.com/leaderboards/lithuanian/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Lithuanian NLU - Atsiliepimai":{"url":"https://euroeval.com/leaderboards/lithuanian/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Lithuanian NLU - MultiWikiQA LT":{"url":"https://euroeval.com/leaderboards/lithuanian/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Lithuanian NLU - ScaLA LT":{"url":"https://euroeval.com/leaderboards/lithuanian/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Lithuanian NLU - WikiANN LT":{"url":"https://euroeval.com/leaderboards/lithuanian/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Lithuanian Summarization - Lrytas":{"url":"https://euroeval.com/leaderboards/lithuanian/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Norwegian":{"url":"https://euroeval.com/leaderboards/norwegian/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Norwegian Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/norwegian/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Norwegian Knowledge":{"url":"https://euroeval.com/leaderboards/norwegian/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Norwegian NLU":{"url":"https://euroeval.com/leaderboards/norwegian/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Norwegian NLU - NorNE NB":{"url":"https://euroeval.com/leaderboards/norwegian/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Norwegian NLU - NorNE NN":{"url":"https://euroeval.com/leaderboards/norwegian/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Norwegian NLU - Norec":{"url":"https://euroeval.com/leaderboards/norwegian/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Norwegian NLU - Norquad":{"url":"https://euroeval.com/leaderboards/norwegian/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Norwegian NLU - ScaLA NB":{"url":"https://euroeval.com/leaderboards/norwegian/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Norwegian NLU - ScaLA NN":{"url":"https://euroeval.com/leaderboards/norwegian/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Norwegian Summarization - NO Sammendrag":{"url":"https://euroeval.com/leaderboards/norwegian/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Polish":{"url":"https://euroeval.com/leaderboards/polish/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Polish Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/polish/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Polish Knowledge":{"url":"https://euroeval.com/leaderboards/polish/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Polish NLU":{"url":"https://euroeval.com/leaderboards/polish/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Polish NLU - KPWr NER":{"url":"https://euroeval.com/leaderboards/polish/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Polish NLU - PoQuAD":{"url":"https://euroeval.com/leaderboards/polish/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Polish NLU - Polemo2":{"url":"https://euroeval.com/leaderboards/polish/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Polish NLU - ScaLA PL":{"url":"https://euroeval.com/leaderboards/polish/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Polish Summarization - PSC":{"url":"https://euroeval.com/leaderboards/polish/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Portuguese":{"url":"https://euroeval.com/leaderboards/portuguese/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Portuguese Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/portuguese/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Portuguese Knowledge":{"url":"https://euroeval.com/leaderboards/portuguese/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Portuguese NLU":{"url":"https://euroeval.com/leaderboards/portuguese/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Portuguese NLU - HAREM":{"url":"https://euroeval.com/leaderboards/portuguese/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Portuguese NLU - MultiWikiQA PT":{"url":"https://euroeval.com/leaderboards/portuguese/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Portuguese NLU - SST-2 PT":{"url":"https://euroeval.com/leaderboards/portuguese/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Portuguese NLU - ScaLA PT":{"url":"https://euroeval.com/leaderboards/portuguese/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Portuguese Summarization - Publico":{"url":"https://euroeval.com/leaderboards/portuguese/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Romanian":{"url":"https://euroeval.com/leaderboards/romanian/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Romanian Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/romanian/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Romanian Knowledge":{"url":"https://euroeval.com/leaderboards/romanian/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Romanian NLU":{"url":"https://euroeval.com/leaderboards/romanian/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Romanian NLU - MultiWikiQA RO":{"url":"https://euroeval.com/leaderboards/romanian/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Romanian NLU - RONEC":{"url":"https://euroeval.com/leaderboards/romanian/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Romanian NLU - RoSent":{"url":"https://euroeval.com/leaderboards/romanian/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Romanian NLU - ScaLA RO":{"url":"https://euroeval.com/leaderboards/romanian/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Romanian Summarization - Sumo RO":{"url":"https://euroeval.com/leaderboards/romanian/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Serbian":{"url":"https://euroeval.com/leaderboards/serbian/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Serbian Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/serbian/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Serbian Knowledge":{"url":"https://euroeval.com/leaderboards/serbian/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Serbian NLU":{"url":"https://euroeval.com/leaderboards/serbian/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Serbian NLU - MMS SR":{"url":"https://euroeval.com/leaderboards/serbian/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Serbian NLU - MultiWikiQA SR":{"url":"https://euroeval.com/leaderboards/serbian/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Serbian NLU - ScaLA SR":{"url":"https://euroeval.com/leaderboards/serbian/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Serbian NLU - UNER SR":{"url":"https://euroeval.com/leaderboards/serbian/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Serbian Summarization - LR SUM SR":{"url":"https://euroeval.com/leaderboards/serbian/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Slovak":{"url":"https://euroeval.com/leaderboards/slovak/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Slovak Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/slovak/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Slovak Knowledge":{"url":"https://euroeval.com/leaderboards/slovak/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Slovak NLU":{"url":"https://euroeval.com/leaderboards/slovak/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Slovak NLU - CSFD Sentiment SK":{"url":"https://euroeval.com/leaderboards/slovak/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Slovak NLU - Multi Wiki QA SK":{"url":"https://euroeval.com/leaderboards/slovak/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Slovak NLU - ScaLA SK":{"url":"https://euroeval.com/leaderboards/slovak/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Slovak NLU - UNER SK":{"url":"https://euroeval.com/leaderboards/slovak/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Slovene":{"url":"https://euroeval.com/leaderboards/slovene/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Slovene Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/slovene/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Slovene Knowledge":{"url":"https://euroeval.com/leaderboards/slovene/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Slovene NLU":{"url":"https://euroeval.com/leaderboards/slovene/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Slovene NLU - MultiWikiQA SL":{"url":"https://euroeval.com/leaderboards/slovene/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Slovene NLU - SSJ500K NER":{"url":"https://euroeval.com/leaderboards/slovene/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Slovene NLU - ScaLA SL":{"url":"https://euroeval.com/leaderboards/slovene/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Slovene NLU - Sentinews":{"url":"https://euroeval.com/leaderboards/slovene/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Spanish":{"url":"https://euroeval.com/leaderboards/spanish/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Spanish Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/spanish/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Spanish Knowledge":{"url":"https://euroeval.com/leaderboards/spanish/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Spanish NLU":{"url":"https://euroeval.com/leaderboards/spanish/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Spanish NLU - CoNLL ES":{"url":"https://euroeval.com/leaderboards/spanish/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Spanish NLU - MLQA ES":{"url":"https://euroeval.com/leaderboards/spanish/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Spanish NLU - ScaLA ES":{"url":"https://euroeval.com/leaderboards/spanish/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Spanish NLU - Sentiment Headlines ES":{"url":"https://euroeval.com/leaderboards/spanish/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Spanish Summarization - Mlsum ES":{"url":"https://euroeval.com/leaderboards/spanish/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Swedish":{"url":"https://euroeval.com/leaderboards/swedish/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Swedish Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/swedish/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Swedish Knowledge":{"url":"https://euroeval.com/leaderboards/swedish/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Swedish NLU":{"url":"https://euroeval.com/leaderboards/swedish/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Swedish NLU - Multi Wiki QA SV":{"url":"https://euroeval.com/leaderboards/swedish/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Swedish NLU - SUC3":{"url":"https://euroeval.com/leaderboards/swedish/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Swedish NLU - ScaLA SV":{"url":"https://euroeval.com/leaderboards/swedish/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Swedish NLU - Swerec":{"url":"https://euroeval.com/leaderboards/swedish/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Swedish Summarization - Swedn":{"url":"https://euroeval.com/leaderboards/swedish/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Ukrainian":{"url":"https://euroeval.com/leaderboards/ukrainian/","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Ukrainian Common Sense Reasoning":{"url":"https://euroeval.com/leaderboards/ukrainian/","metric":"Common Sense Reasoning Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Ukrainian Knowledge":{"url":"https://euroeval.com/leaderboards/ukrainian/","metric":"Knowledge Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Ukrainian NLU":{"url":"https://euroeval.com/leaderboards/ukrainian/","metric":"NLU Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Ukrainian NLU - Cross Domain UK Reviews":{"url":"https://euroeval.com/leaderboards/ukrainian/","metric":"Sentiment classification Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Ukrainian NLU - MultiWikiQA UK":{"url":"https://euroeval.com/leaderboards/ukrainian/","metric":"Reading comprehension Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Ukrainian NLU - NER UK":{"url":"https://euroeval.com/leaderboards/ukrainian/","metric":"Named entity recognition Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EuroEval Ukrainian NLU - ScaLA UK":{"url":"https://euroeval.com/leaderboards/ukrainian/","metric":"Linguistic acceptability Score (%)","theoretical_min":null,"theoretical_max":100.0},"EuroEval Ukrainian Summarization - LR SUM UK":{"url":"https://euroeval.com/leaderboards/ukrainian/","metric":"Score (%)","theoretical_min":null,"theoretical_max":100.0},"European LLM Leaderboard - Accuracy":{"url":"https://huggingface.co/spaces/Eurolingua/european-llm-leaderboard","metric":"Average Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"European LLM Leaderboard - FLORES200":{"url":"https://huggingface.co/spaces/Eurolingua/european-llm-leaderboard","metric":"Average FLORES200 Score","theoretical_min":0.0,"theoretical_max":100.0},"European LLM Leaderboard - FLORES200 Source":{"url":"https://huggingface.co/spaces/Eurolingua/european-llm-leaderboard","metric":"FLORES200 Source Score","theoretical_min":0.0,"theoretical_max":100.0},"European LLM Leaderboard - FLORES200 Target":{"url":"https://huggingface.co/spaces/Eurolingua/european-llm-leaderboard","metric":"FLORES200 Target Score","theoretical_min":0.0,"theoretical_max":100.0},"European LLM Leaderboard - Zero-Shot Accuracy":{"url":"https://huggingface.co/spaces/Eurolingua/european-llm-leaderboard","metric":"Average Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Eval Anything":{"url":"https://huggingface.co/spaces/PKU-Alignment/EvalAnything-LeaderBoard","metric":"Overall score","theoretical_min":0.0,"theoretical_max":null},"EvalPlus":{"url":"https://benchmarklist.com/benchmarks/evalplus/","metric":"EvalPlus Avg. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"EvalPlus (HumanEval+ & MBPP+)":{"url":"https://evalplus.github.io/leaderboard.html","metric":"Pass@1 avg (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - ARC":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Classification":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ab":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ace":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ach":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language aeb":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language af":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ak":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language am":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language an":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language apc":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ar":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ars":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ary":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language arz":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language as":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ast":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language awa":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ay":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language az":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ba":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ban":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language bbc":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language be":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language bem":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language bew":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language bg":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language bho":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language bik":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language bjn":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language bm":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language bn":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language bo":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language br":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language brx":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language bs":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language bua":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language bug":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ca":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ceb":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language cgg":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language chm":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ckb":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language co":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language crh":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language crs":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language cs":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language cv":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language cy":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language da":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language dar":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language de":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language doi":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language dv":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language dyu":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language dz":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ee":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language el":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language en":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language eo":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language es":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language et":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language eu":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language fa":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ff":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language fi":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language fil":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language fj":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language fo":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language fon":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language fr":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language fur":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language fuv":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language fy":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ga":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language gaa":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language gd":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language gl":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language gn":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language gom":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language gu":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ha":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language haw":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language he":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language hi":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language hil":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language hne":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language hr":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ht":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language hu":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language hy":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language id":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ig":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ilo":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language is":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language it":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ja":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language jv":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ka":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language kab":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language kac":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language kam":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language kea":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ki":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language kk":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language km":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language kmb":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language kn":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ko":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language kri":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ks":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ku":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ky":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language la":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language lb":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language lg":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language li":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language lij":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language lmo":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ln":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language lo":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language lt":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ltg":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language lua":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language luo":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language lv":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language mag":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language mai":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language mak":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language mfe":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language mg":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language mi":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language min":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language mk":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ml":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language mn":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language mni":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language mos":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language mr":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ms":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language mt":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language my":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language myv":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language nb":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ne":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language new":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language nl":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language nn":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language nqo":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language nr":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language nso":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language nus":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ny":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language oc":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language om":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language or":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language pa":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language pag":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language pam":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language pap":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language pl":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ps":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language pt":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language qu":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language rn":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ro":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ru":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language rw":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language sa":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language sat":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language sc":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language scn":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language sd":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language sg":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language shn":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language si":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language sk":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language sl":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language sm":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language sn":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language so":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language sq":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language sr":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ss":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language st":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language su":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language sv":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language sw":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language szl":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ta":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language te":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language tet":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language tg":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language th":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ti":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language tk":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language tn":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language tpi":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language tr":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ts":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language tt":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language tum":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language tyv":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ug":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language uk":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language umb":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language ur":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language uz":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language vai":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language vec":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language vi":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language vmw":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language war":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language wo":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language wuu":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language xh":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language yi":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language yo":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language yua":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language yue":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language zgh":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language zh":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Language zu":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - MGSM":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - MMLU":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Translation From":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - Translation To":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Evals for Every Language - TruthfulQA":{"url":"https://huggingface.co/spaces/fair-forward/evals-for-every-language","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"EvasionBench":{"url":"https://iiiiqiiii.github.io/EvasionBench/","metric":"Macro-F1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"Every Act Has Its Price":{"url":"https://benchmarklist.com/benchmarks/every_act_has_its_price_compressed_moral_composition_in_frontier_llms/","metric":"Care (self-reported)","theoretical_min":0.0,"theoretical_max":null},"EvoCode-Bench":{"url":"https://benchmarklist.com/benchmarks/evocode_bench/","metric":"MT@4 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"EvoEval":{"url":"https://evo-eval.github.io/leaderboard.html","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"EvoEval Combine":{"url":"https://evo-eval.github.io/leaderboard.html","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"EvoEval Concise":{"url":"https://evo-eval.github.io/leaderboard.html","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"EvoEval Creative":{"url":"https://evo-eval.github.io/leaderboard.html","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"EvoEval Difficult":{"url":"https://evo-eval.github.io/leaderboard.html","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"EvoEval Subtle":{"url":"https://evo-eval.github.io/leaderboard.html","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"EvoEval Tool Use":{"url":"https://evo-eval.github.io/leaderboard.html","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"EvoEval Verbose":{"url":"https://evo-eval.github.io/leaderboard.html","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"ExploitBench v8-bench":{"url":"https://exploitbench.ai/","metric":"Mean Capability (%)","theoretical_min":0.0,"theoretical_max":100.0},"ExploitGym":{"url":"https://www.cybergym.io/exploitgym/","metric":"Successful Intended Exploits (#)","theoretical_min":0.0,"theoretical_max":null},"Exploring Autonomous Agentic Data Engineering":{"url":"https://benchmarklist.com/benchmarks/exploring_autonomous_agentic_data_engineering_for_model_specialization/","metric":"MATS Science (Scratch) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"FACTS Leaderboard":{"url":"https://huggingface.co/spaces/MaziyarPanahi/FACTS-Leaderboard","metric":"Combined Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"FAM-Bench":{"url":"https://benchmarklist.com/benchmarks/fam_bench/","metric":"Task 1 Acc. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"FAVOR-Bench":{"url":"https://favor-bench.github.io/","metric":"Overall Score","theoretical_min":0.0,"theoretical_max":100.0},"FINAL Bench Metacognitive":{"url":"https://benchmarklist.com/benchmarks/final_bench_metacognitive/","metric":"Metacognitive Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"FIND":{"url":"https://benchmarklist.com/benchmarks/find_toward_multimodal_financial_reasoning_and_question_answering_for_indic_languages/","metric":"Zero shot en (self-reported)","theoretical_min":0.0,"theoretical_max":null},"FINESSE-Bench":{"url":"https://benchmarklist.com/benchmarks/finesse_bench_a_hierarchical_benchmark_suite_for_financial_domain_knowledge_and_technical_analysis_in_large_language_models/","metric":"exam-like (self-reported)","theoretical_min":0.0,"theoretical_max":null},"FLORES European Languages Leaderboard":{"url":"https://huggingface.co/spaces/speakleash/leaderboard-flores","metric":"Average chrF","theoretical_min":0.0,"theoretical_max":100.0},"FLORES European Languages Leaderboard - BLEU":{"url":"https://huggingface.co/spaces/speakleash/leaderboard-flores","metric":"Average BLEU","theoretical_min":0.0,"theoretical_max":100.0},"FMNB Leaderboard":{"url":"https://huggingface.co/spaces/TheFey/MNB-Leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"FORTIS":{"url":"https://benchmarklist.com/benchmarks/fortis_benchmarking_over_privilege_in_agent_skills/","metric":"Task 1 EM (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Factorio Learning Environment":{"url":"https://jackhopkins.github.io/factorio-learning-environment/leaderboard/","metric":"Production Score","theoretical_min":null,"theoretical_max":null},"Factory Code Review Benchmark":{"url":"https://factory.ai/news/code-review-benchmark","metric":"Mean F1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"FactoryBench":{"url":"https://benchmarklist.com/benchmarks/factorybench/","metric":"Overall Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"FanOutQA":{"url":"https://fanoutqa.com/leaderboard","metric":"Accuracy (%, closedbook loose)","theoretical_min":0.0,"theoretical_max":100.0},"FastEval":{"url":"https://fasteval.github.io/FastEval/","metric":"Total Score","theoretical_min":0.0,"theoretical_max":100.0},"Fibble Arena":{"url":"https://drchangliu.github.io/WordleArenas/","metric":"Win Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Fibble2 Arena":{"url":"https://drchangliu.github.io/WordleArenas/","metric":"Win Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Fibble3 Arena":{"url":"https://drchangliu.github.io/WordleArenas/","metric":"Win Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Fibble4 Arena":{"url":"https://drchangliu.github.io/WordleArenas/","metric":"Win Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Fibble5 Arena":{"url":"https://drchangliu.github.io/WordleArenas/","metric":"Win Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Fiction.LiveBench":{"url":"https://fiction.live/stories/Fiction-liveBench-Mar-14-2025/oQdzQvKHw8JyXbN87/home","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"FigQA":{"url":"https://benchmarklist.com/benchmarks/figqa/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Fin-Bias":{"url":"https://benchmarklist.com/benchmarks/fin_bias_comprehensive_evaluation_for_llm_decision_making_under_human_bias_in_finance_domain/","metric":"Average Herding Score (with rating) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Fin-RATE":{"url":"https://benchmarklist.com/benchmarks/fin_rate/","metric":"Macro Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"FinBen (Financial LLM)":{"url":"https://huggingface.co/spaces/TheFinAI/Open-FinLLM-Leaderboard","metric":"Average Score","theoretical_min":0.0,"theoretical_max":100.0},"FinBen - FNS":{"url":"https://huggingface.co/spaces/TheFinAI/Open-FinLLM-Leaderboard","metric":"Normalized Score","theoretical_min":0.0,"theoretical_max":100.0},"FinBen - FinNum":{"url":"https://huggingface.co/spaces/TheFinAI/Open-FinLLM-Leaderboard","metric":"Normalized Score","theoretical_min":0.0,"theoretical_max":100.0},"FinBen - FinText":{"url":"https://huggingface.co/spaces/TheFinAI/Open-Financial-LLM-Leaderboard","metric":"Normalized Score","theoretical_min":0.0,"theoretical_max":100.0},"FinBen - MultiFin":{"url":"https://huggingface.co/spaces/TheFinAI/Open-FinLLM-Leaderboard","metric":"Normalized Score","theoretical_min":0.0,"theoretical_max":100.0},"FinBen - QA":{"url":"https://huggingface.co/spaces/TheFinAI/Open-FinLLM-Leaderboard","metric":"Normalized Score","theoretical_min":0.0,"theoretical_max":100.0},"FinChain":{"url":"https://benchmarklist.com/benchmarks/finchain/","metric":"ChainEval (self-reported)","theoretical_min":0.0,"theoretical_max":null},"FinSearchComp":{"url":"https://randomtutu.github.io/FinSearchComp/","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"FinVerBench":{"url":"https://benchmarklist.com/benchmarks/finverbench/","metric":"Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Finance Agent (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Finance Agent v1.1":{"url":"https://benchmarklist.com/benchmarks/vals_finance_agent/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Finance Agent v2":{"url":"https://benchmarklist.com/benchmarks/vals_finance_agent_v2/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"FinanceArena":{"url":"https://benchmarklist.com/benchmarks/financearena/","metric":"Overall Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Finetuning with Scientific Data Increases Hall":{"url":"https://benchmarklist.com/benchmarks/finetuning_with_scientific_data_increases_hallucinations_a_multi_domain_factuality_evaluation_of_llms/","metric":"OFS (self-reported)","theoretical_min":0.0,"theoretical_max":null},"FoodTruckBench":{"url":"https://foodtruckbench.com","metric":"Net Worth ($)","theoretical_min":null,"theoretical_max":null},"ForecastBench":{"url":"https://www.forecastbench.org/tournament/","metric":"Overall Score (higher is better)","theoretical_min":0.0,"theoretical_max":100.0},"Foresight Bench":{"url":"https://benchmark.aleatoricai.com/leaderboard","metric":"Forecast Skill (100 - Brier x 100)","theoretical_min":0.0,"theoretical_max":100.0},"Forge":{"url":"https://benchmarklist.com/benchmarks/forge_quality_aware_reinforcement_learning_for_np_hard_optimization_in_llms/","metric":"Overall SR (self-reported)","theoretical_min":0.0,"theoretical_max":null},"FormInv":{"url":"https://benchmarklist.com/benchmarks/forminv/","metric":"SCR (self-reported)","theoretical_min":0.0,"theoretical_max":null},"FormalRewardBench":{"url":"https://benchmarklist.com/benchmarks/formalrewardbench_a_benchmark_for_formal_theorem_proving_reward_models/","metric":"Pairwise Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"FormationEval":{"url":"https://huggingface.co/datasets/AlmazErmilov/FormationEval","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"FraudBench":{"url":"https://benchmarklist.com/benchmarks/fraudbench/","metric":"GPT Image 2 TPR (self-reported)","theoretical_min":0.0,"theoretical_max":null},"French LLM Leaderboard - Average":{"url":"https://huggingface.co/spaces/fr-gouv-coordination-ia/llm_leaderboard_fr","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"French LLM Leaderboard - BAC FR":{"url":"https://huggingface.co/spaces/fr-gouv-coordination-ia/llm_leaderboard_fr","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"French LLM Leaderboard - GPQA FR":{"url":"https://huggingface.co/spaces/fr-gouv-coordination-ia/llm_leaderboard_fr","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"French LLM Leaderboard - IFEval FR":{"url":"https://huggingface.co/spaces/fr-gouv-coordination-ia/llm_leaderboard_fr","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"From 0-Order Selection to 2-Order Judgment":{"url":"https://benchmarklist.com/benchmarks/from_0_order_selection_to_2_order_judgment_combinatorial_hardening_exposes_compositional_failures_in_frontier_llms/","metric":"H-Comb (self-reported)","theoretical_min":0.0,"theoretical_max":null},"From Hallucination to Grounding":{"url":"https://benchmarklist.com/benchmarks/from_hallucination_to_grounding_diagnosing_visual_spatial_intelligence_via_crisp/","metric":"Con. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"From Knowing to Doing":{"url":"https://benchmarklist.com/benchmarks/from_knowing_to_doing_a_memory_controlled_benchmark_for_llm_trading_agents_on_stock_markets/","metric":"Total ret. (self-reported)","theoretical_min":null,"theoretical_max":null},"From Perception to Action":{"url":"https://benchmarklist.com/benchmarks/perception_to_action/","metric":"pass@1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Frontier Security Agents - BOTS v1":{"url":"https://arxiv.org/html/2607.15263v1","metric":"BOTS score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Frontier Security Agents - Cybench Hard":{"url":"https://arxiv.org/html/2607.15263v1","metric":"Success rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"FrontierCode":{"url":"https://benchmarklist.com/benchmarks/frontiercode/","metric":"Main Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"FrontierCode Diamond":{"url":"https://cognition.ai/blog/frontier-code","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"FrontierCode Diamond (Fable/Mythos)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"FrontierCode Extended":{"url":"https://cognition.ai/blog/frontier-code","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"FrontierCode Main":{"url":"https://cognition.ai/blog/frontier-code","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"FrontierMath - Tier 4":{"url":"https://epoch.ai/frontiermath/tiers-1-4","metric":"Accuracy (%, 48 problems)","theoretical_min":0.0,"theoretical_max":100.0},"FrontierMath - Tier 4 (v2)":{"url":"https://epoch.ai/benchmarks/frontiermath-tier-4-v2","metric":"Accuracy (%, 41 private v2 problems)","theoretical_min":0.0,"theoretical_max":100.0},"FrontierMath - Tiers 1-3":{"url":"https://epoch.ai/frontiermath/tiers-1-4","metric":"Accuracy (%, 290 problems)","theoretical_min":0.0,"theoretical_max":100.0},"FrontierMath - Tiers 1-3 (v2)":{"url":"https://epoch.ai/benchmarks/frontiermath-tiers-1-3-v2","metric":"Accuracy (%, 285 private v2 problems)","theoretical_min":0.0,"theoretical_max":100.0},"FrontierOR":{"url":"https://benchmarklist.com/benchmarks/frontieror/","metric":"Sol. quality (self-reported)","theoretical_min":0.0,"theoretical_max":null},"FrontierSWE":{"url":"https://www.frontierswe.com/","metric":"Dominance (%)","theoretical_min":0.0,"theoretical_max":100.0},"FullHome":{"url":"https://benchmarklist.com/benchmarks/fullhome/","metric":"Goal SR (VirtualHome) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"FullStackBench en":{"url":"https://benchmarklist.com/benchmarks/fullstackbench_en/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"FutureEval":{"url":"https://www.metaculus.com/futureeval/leaderboard/","metric":"Unified Forecasting Score","theoretical_min":null,"theoretical_max":null},"FutureSearch BTF-2":{"url":"https://evals.futuresearch.ai/#btf","metric":"Brier Score","theoretical_min":0.0,"theoretical_max":1.0},"FutureSearch BTF-2 Calibration":{"url":"https://evals.futuresearch.ai/#btf","metric":"Calibration Error","theoretical_min":0.0,"theoretical_max":1.0},"FutureSearch BTF-2 Refinement":{"url":"https://evals.futuresearch.ai/#btf","metric":"Refinement","theoretical_min":0.0,"theoretical_max":1.0},"FutureSearch DRB":{"url":"https://evals.futuresearch.ai/#drb","metric":"Average Score","theoretical_min":0.0,"theoretical_max":1.0},"FutureSearch DRB - Compile Dataset":{"url":"https://evals.futuresearch.ai/#drb","metric":"Average Score","theoretical_min":0.0,"theoretical_max":1.0},"FutureSearch DRB - Derive Number":{"url":"https://evals.futuresearch.ai/#drb","metric":"Average Score","theoretical_min":0.0,"theoretical_max":1.0},"FutureSearch DRB - Find Dataset":{"url":"https://evals.futuresearch.ai/#drb","metric":"Average Score","theoretical_min":0.0,"theoretical_max":1.0},"FutureSearch DRB - Find Number":{"url":"https://evals.futuresearch.ai/#drb","metric":"Average Score","theoretical_min":0.0,"theoretical_max":1.0},"FutureSearch DRB - Find Original Source":{"url":"https://evals.futuresearch.ai/#drb","metric":"Average Score","theoretical_min":0.0,"theoretical_max":1.0},"FutureSearch DRB - Gather Evidence":{"url":"https://evals.futuresearch.ai/#drb","metric":"Average Score","theoretical_min":0.0,"theoretical_max":1.0},"FutureSearch DRB - Populate Reference Class":{"url":"https://evals.futuresearch.ai/#drb","metric":"Average Score","theoretical_min":0.0,"theoretical_max":1.0},"FutureSearch DRB - Validate Claim":{"url":"https://evals.futuresearch.ai/#drb","metric":"Average Score","theoretical_min":0.0,"theoretical_max":1.0},"FutureSim":{"url":"https://openforecaster.github.io/futuresim/","metric":"Top-1 Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"FutureX":{"url":"https://futurex.live/","metric":"Overall Score (latest week, %)","theoretical_min":0.0,"theoretical_max":100.0},"GACL - 2x8 MiniChess":{"url":"https://gameagentcodingleague.com/leaderboard.html","metric":"Normalized Score (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"GACL - Battleship":{"url":"https://gameagentcodingleague.com/leaderboard.html","metric":"Normalized Score (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"GACL - Connect4":{"url":"https://gameagentcodingleague.com/leaderboard.html","metric":"Normalized Score (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"GACL - Surround Morris":{"url":"https://gameagentcodingleague.com/leaderboard.html","metric":"Normalized Score (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"GACL - Tic-Tac-Toe":{"url":"https://gameagentcodingleague.com/leaderboard.html","metric":"Normalized Score (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"GACL - Wizard":{"url":"https://gameagentcodingleague.com/leaderboard.html","metric":"Normalized Score (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"GACL - WordMatrix":{"url":"https://gameagentcodingleague.com/leaderboard.html","metric":"Normalized Score (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"GAIA":{"url":"https://huggingface.co/spaces/gaia-benchmark/leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"GAIA2":{"url":"https://huggingface.co/spaces/meta-agents-research-environments/leaderboard","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"GAIA2 - Adaptability":{"url":"https://huggingface.co/spaces/meta-agents-research-environments/leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"GAIA2 - Ambiguity":{"url":"https://huggingface.co/spaces/meta-agents-research-environments/leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"GAIA2 - Execution":{"url":"https://huggingface.co/spaces/meta-agents-research-environments/leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"GAIA2 - Noise":{"url":"https://huggingface.co/spaces/meta-agents-research-environments/leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"GAIA2 - Search":{"url":"https://huggingface.co/spaces/meta-agents-research-environments/leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"GBA Eval":{"url":"https://gbaeval.com/","metric":"Overall Emulator Score","theoretical_min":0.0,"theoretical_max":null},"GDM InterCode CTF":{"url":"https://github.com/UKGovernmentBEIS/inspect_evals/tree/main/src/inspect_evals/gdm_intercode_ctf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"GDP.pdf":{"url":"https://surgehq.ai/leaderboards","metric":"Strict Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"GDP.pdf (Fable/Mythos)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Strict Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"GDPval":{"url":"https://benchmarklist.com/benchmarks/openai_gdpval/","metric":"Wins/Ties vs Human (self-reported)","theoretical_min":0.0,"theoretical_max":null},"GDPval (OpenAI Evals)":{"url":"https://evals.openai.com/gdpval/leaderboard","metric":"Win Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"GDPval-AA":{"url":"https://artificialanalysis.ai/evaluations/gdpval-aa","metric":"Elo","theoretical_min":null,"theoretical_max":null},"GDPval-AA (Fable/Mythos)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Elo","theoretical_min":null,"theoretical_max":null},"GDPval-MM":{"url":"https://benchmarklist.com/benchmarks/gdpval_mm/","metric":"Wins + ties (self-reported)","theoretical_min":0.0,"theoretical_max":null},"GENSTRAT":{"url":"https://benchmarklist.com/benchmarks/genstrat_toward_a_science_of_strategic_reasoning_in_large_language_models/","metric":"Alpha (chips/game) (self-reported)","theoretical_min":null,"theoretical_max":null},"GIM":{"url":"https://arxiv.org/abs/2605.18663","metric":"IRT ability (theta)","theoretical_min":null,"theoretical_max":null},"GISA":{"url":"https://huggingface.co/spaces/RUC-NLPIR/GISA-LeaderBoard","metric":"Overall Score","theoretical_min":0.0,"theoretical_max":100.0},"GMMLU (Anthropic)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Average Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"GPQA Diamond (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"GQA":{"url":"https://eval.ai/web/challenges/challenge-page/2257/leaderboard","metric":"Overall Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"GR-Ben":{"url":"https://benchmarklist.com/benchmarks/gr_ben_a_general_reasoning_benchmark_for_evaluating_process_reward_models/","metric":"Average (self-reported)","theoretical_min":0.0,"theoretical_max":null},"GRAB-Lite":{"url":"https://grab-benchmark.github.io/","metric":"Overall Score","theoretical_min":0.0,"theoretical_max":100.0},"GRACE":{"url":"https://benchmarklist.com/benchmarks/grace/","metric":"Avg Step (self-reported)","theoretical_min":0.0,"theoretical_max":null},"GRIPS":{"url":"https://huggingface.co/datasets/ellamind/grips","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"GRIPS-hard":{"url":"https://huggingface.co/datasets/ellamind/grips","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"GSM-MC":{"url":"https://github.com/Geralt-Targaryen/MC-Evaluation","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"GSM8K":{"url":"https://epoch.ai/benchmarks/gsm8k","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"GSMA Open-Telco - 3GPP":{"url":"https://huggingface.co/spaces/GSMA/open-telco-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"GSMA Open-Telco - ORAN-Bench":{"url":"https://huggingface.co/spaces/GSMA/open-telco-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"GSMA Open-Telco - TeleLogs":{"url":"https://huggingface.co/spaces/GSMA/open-telco-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"GSMA Open-Telco - TeleMath":{"url":"https://huggingface.co/spaces/GSMA/open-telco-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"GSMA Open-Telco - TeleQnA":{"url":"https://huggingface.co/spaces/GSMA/open-telco-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"GSMA Open-Telco - TeleTables":{"url":"https://huggingface.co/spaces/GSMA/open-telco-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"GSMA Open-Telco - srsRAN-Bench":{"url":"https://huggingface.co/spaces/GSMA/open-telco-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"GSMA Open-Telco LLM Leaderboard":{"url":"https://huggingface.co/spaces/GSMA/open-telco-llm-leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"GSO-Bench":{"url":"https://gso-bench.github.io/","metric":"Opt@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"GTA":{"url":"https://github.com/open-compass/GTA","metric":"Answer Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"GUI vs. CLI":{"url":"https://benchmarklist.com/benchmarks/gui_vs_cli_execution_bottlenecks_in_screen_only_and_skill_mediated_computer_use_agents/","metric":"Avg. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Galileo Agent - Banking Accuracy":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Agent - Banking TSQ":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Task Success Quality (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Agent - Healthcare Accuracy":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Agent - Healthcare TSQ":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Task Success Quality (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Agent - Insurance Accuracy":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Agent - Insurance TSQ":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Task Success Quality (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Agent - Investment Accuracy":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Agent - Investment TSQ":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Task Success Quality (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Agent - Telecom Accuracy":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Agent - Telecom TSQ":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Task Success Quality (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Agent Leaderboard":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Avg Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Tool Tasks - BFCL v3 Irrelevance":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Tool Tasks - BFCL v3 Multi-Turn Base Multi-Function":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Tool Tasks - BFCL v3 Multi-Turn Base Single Function":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Tool Tasks - BFCL v3 Multi-Turn Composite":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Tool Tasks - BFCL v3 Multi-Turn Long Context":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Tool Tasks - BFCL v3 Multi-Turn Missing Function":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Tool Tasks - BFCL v3 Multi-Turn Missing Parameter":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Tool Tasks - Multi Turn":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Tool Tasks - Single Turn":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Tool Tasks - ToolACE Single Function Call":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Tool Tasks - tau Long Context":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Tool Tasks - xLAM Multiple Tool Multiple Call":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Tool Tasks - xLAM Multiple Tool Single Call":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Tool Tasks - xLAM Single Tool Multiple Call":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Tool Tasks - xLAM Single Tool Single Call":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Galileo Tool Tasks - xLAM Tool Missing":{"url":"https://huggingface.co/spaces/galileo-ai/agent-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Game Agent Coding League":{"url":"https://gameagentcodingleague.com/leaderboard.html","metric":"Weighted Score (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"GameCraft-Bench":{"url":"https://tongxuluo.github.io/gamecraft-bench-website/","metric":"Overall Score","theoretical_min":null,"theoretical_max":null},"GameWorld Computer-Use":{"url":"https://gameworld-project.github.io/","metric":"Progress (%)","theoretical_min":0.0,"theoretical_max":100.0},"GameWorld Generalist":{"url":"https://gameworld-project.github.io/","metric":"Progress (%)","theoretical_min":0.0,"theoretical_max":100.0},"Gapminder AI Worldview":{"url":"https://www.gapminder.org/ai/worldview_benchmark/","metric":"Correct Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"GeneBench":{"url":"https://benchmarklist.com/benchmarks/genebench/","metric":"Mean pass rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Generalization V1 (Lechmazur)":{"url":"https://github.com/lechmazur/generalization/tree/main/v1","metric":"Avg Rank (lower is better)","theoretical_min":1.0,"theoretical_max":8.0},"Generalization V2 (Lechmazur)":{"url":"https://github.com/lechmazur/generalization","metric":"Inverse-Rank Score","theoretical_min":0.0,"theoretical_max":100.0},"Generate README Eval":{"url":"https://benchmarklist.com/benchmarks/generate_readme_eval/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Geo-Eval":{"url":"https://benchmarklist.com/benchmarks/geo_eval/","metric":"Average (self-reported)","theoretical_min":0.0,"theoretical_max":null},"GeoBench ACW":{"url":"https://geobench.org/","metric":"Average Score","theoretical_min":0.0,"theoretical_max":5000.0},"GeoBench Photos":{"url":"https://geobench.org/","metric":"Average Score","theoretical_min":0.0,"theoretical_max":5000.0},"GeoCode Leaderboard":{"url":"https://benchmarklist.com/benchmarks/geocode_leaderboard/","metric":"AutoGEEval++ Pass@1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"GeoDial":{"url":"https://benchmarklist.com/benchmarks/geodial_a_multimodal_conversational_tutoring_dataset_for_geometry_problem_solving_with_visual_tutor_turns/","metric":"Act Macro-F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"GeoRC":{"url":"https://benchmarklist.com/benchmarks/georc/","metric":"F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"GeoSym-Bench":{"url":"https://benchmarklist.com/benchmarks/geosym_bench/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Gert Labs Rankings":{"url":"https://gertlabs.com/","metric":"GScore (%)","theoretical_min":0.0,"theoretical_max":100.0},"GlobalDentBench":{"url":"https://benchmarklist.com/benchmarks/globaldentbench/","metric":"Macro-Average Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Google Gemini 3 Deep Think - ARC-AGI-2":{"url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_3_deep_think_model_evaluation.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Google Gemini 3 Deep Think - CMT-Benchmark":{"url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_3_deep_think_model_evaluation.pdf","metric":"Pass@8 (%)","theoretical_min":0.0,"theoretical_max":100.0},"Google Gemini 3 Deep Think - Codeforces":{"url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_3_deep_think_model_evaluation.pdf","metric":"Elo","theoretical_min":null,"theoretical_max":null},"Google Gemini 3 Deep Think - GPQA Diamond":{"url":"https://blog.google/products-and-platforms/products/gemini/gemini-3/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Google Gemini 3 Deep Think - Humanity's Last Exam (no tools)":{"url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_3_deep_think_model_evaluation.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Google Gemini 3 Deep Think - Humanity's Last Exam (search and code)":{"url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_3_deep_think_model_evaluation.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Google Gemini 3 Deep Think - International Chemistry Olympiad 2025 (theory)":{"url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_3_deep_think_model_evaluation.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Google Gemini 3 Deep Think - International Math Olympiad 2025":{"url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_3_deep_think_model_evaluation.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Google Gemini 3 Deep Think - International Physics Olympiad 2025 (theory)":{"url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_3_deep_think_model_evaluation.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Google Gemini 3 Deep Think - MMMU-Pro":{"url":"https://storage.googleapis.com/deepmind-media/gemini/gemini_3_deep_think_model_evaluation.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Gorilla API Bench (BFCL)":{"url":"https://gorilla.cs.berkeley.edu/leaderboard.html","metric":"Overall Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"GosuEvals":{"url":"https://gosuevals.com/leaderboard.html","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"GraphARC":{"url":"https://benchmarklist.com/benchmarks/grapharc/","metric":"Average accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"GraphWalks BFS 256K (Anthropic)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"F1 Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"GraphWalks BFS 256K (Fable/Mythos)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"GraphWalks Parents 256K (Anthropic)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"F1 Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"GraphWalks Parents 256K (Fable/Mythos)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Graphwalks BFS 1M F1":{"url":"https://benchmarklist.com/benchmarks/graphwalks_bfs_1m_f1/","metric":"F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Graphwalks BFS 256k F1":{"url":"https://benchmarklist.com/benchmarks/graphwalks_bfs_256k_f1/","metric":"F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Graphwalks Parents 1M F1":{"url":"https://benchmarklist.com/benchmarks/graphwalks_parents_1m_f1/","metric":"F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Graphwalks Parents 256k F1":{"url":"https://benchmarklist.com/benchmarks/graphwalks_parents_256k_f1/","metric":"F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Greek MMLU":{"url":"https://huggingface.co/spaces/yangzhang33/GreekMMLU-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Greek MMLU - Greek-Specific":{"url":"https://huggingface.co/spaces/yangzhang33/GreekMMLU-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Greek MMLU - Humanities":{"url":"https://huggingface.co/spaces/yangzhang33/GreekMMLU-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Greek MMLU - Other":{"url":"https://huggingface.co/spaces/yangzhang33/GreekMMLU-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Greek MMLU - STEM":{"url":"https://huggingface.co/spaces/yangzhang33/GreekMMLU-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Greek MMLU - Social Sciences":{"url":"https://huggingface.co/spaces/yangzhang33/GreekMMLU-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Greenland Sovereignty Game (implicitly as a st":{"url":"https://benchmarklist.com/benchmarks/greenland_sovereignty_game_implicitly_as_a_structural_benchmark_for_llm_geopolitical_behavior/","metric":"Escalation Composite (self-reported)","theoretical_min":0.0,"theoretical_max":null},"GroupTravelBench":{"url":"https://benchmarklist.com/benchmarks/grouptravelbench/","metric":"GU (Group Utility) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Guesswork":{"url":"https://aibenchmarks.dev/prediction-duel","metric":"MAE (z-score units)","theoretical_min":0.0,"theoretical_max":null},"HAL AssistantBench":{"url":"https://hal.cs.princeton.edu/assistantbench","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HAL CORE-Bench Hard":{"url":"https://hal.cs.princeton.edu/corebench_hard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HAL GAIA":{"url":"https://hal.cs.princeton.edu/gaia","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HAL GAIA Level 1":{"url":"https://hal.cs.princeton.edu/gaia","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HAL GAIA Level 2":{"url":"https://hal.cs.princeton.edu/gaia","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HAL GAIA Level 3":{"url":"https://hal.cs.princeton.edu/gaia","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HAL Online Mind2Web":{"url":"https://hal.cs.princeton.edu/online_mind2web","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HAL SWE-bench Verified Mini":{"url":"https://www.swebench.com","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"HAL SciCode":{"url":"https://hal.cs.princeton.edu/scicode","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HAL ScienceAgentBench":{"url":"https://hal.cs.princeton.edu/scienceagentbench","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HAL TAU-bench Airline":{"url":"https://hal.cs.princeton.edu/taubench_airline","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HAL USACO":{"url":"https://hal.cs.princeton.edu/usaco","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM":{"url":"https://benchmarklist.com/benchmarks/helm/","metric":"Mean win rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"HELM (Stanford)":{"url":"https://crfm.stanford.edu/helm/","metric":"Mean Win Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM AIR-Bench":{"url":"https://storage.googleapis.com/crfm-helm-public/air-bench/benchmark_output/releases/v1.19.0/groups/safety_scenarios.json","metric":"Refusal Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Capabilities - GPQA":{"url":"https://crfm.stanford.edu/helm/capabilities/latest/?group=gpqa","metric":"COT correct","theoretical_min":0.0,"theoretical_max":100.0},"HELM Capabilities - IFEval":{"url":"https://crfm.stanford.edu/helm/capabilities/latest/?group=ifeval","metric":"IFEval Strict Acc","theoretical_min":0.0,"theoretical_max":100.0},"HELM Capabilities - MMLU-Pro":{"url":"https://crfm.stanford.edu/helm/capabilities/latest/?group=mmlu_pro","metric":"COT correct","theoretical_min":0.0,"theoretical_max":100.0},"HELM Capabilities - Omni-MATH":{"url":"https://crfm.stanford.edu/helm/capabilities/latest/?group=omni_math","metric":"Acc","theoretical_min":0.0,"theoretical_max":100.0},"HELM Capabilities - WildBench":{"url":"https://crfm.stanford.edu/helm/capabilities/latest/?group=wildbench","metric":"WB Score","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - BBQ":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=bbq","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - BLiMP":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=blimp","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - BoolQ":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=boolq","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - CNN/DailyMail":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=summarization_cnndm","metric":"ROUGE-2 (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - CivilComments":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=civil_comments","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - Dyck":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=dyck_language","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - Entity Data Imputation":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=entity_data_imputation","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - Entity Matching":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=entity_matching","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - GSM8K":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=gsm","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - HellaSwag":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=hellaswag","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - IMDB":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=imdb","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - LSAT":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=lsat_qa","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - LegalSupport":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=legal_support","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - MATH":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=math_regular","metric":"Equivalent (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - MATH Chain-of-Thought":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=math_chain_of_thought","metric":"Equivalent (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - MMLU":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=mmlu","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - MS MARCO Regular":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=msmarco_regular","metric":"RR@10 (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - MS MARCO TREC":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=msmarco_trec","metric":"NDCG@10 (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - NarrativeQA":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=narrative_qa","metric":"F1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - NaturalQuestions Closed Book":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=natural_qa_closedbook","metric":"F1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - NaturalQuestions Open Book":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=natural_qa_openbook_longans","metric":"F1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - OpenbookQA":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=openbookqa","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - QuAC":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=quac","metric":"F1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - RAFT":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=raft","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - Synthetic Reasoning Abstract":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=synthetic_reasoning","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - Synthetic Reasoning Natural":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=synthetic_reasoning_natural","metric":"F1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - TruthfulQA":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=truthful_qa","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - WikiFact":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=wikifact","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - XSUM":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=summarization_xsum","metric":"ROUGE-2 (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Classic - bAbI":{"url":"https://crfm.stanford.edu/helm/classic/latest/?group=babi_qa","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM EWoK - Agent Properties":{"url":"https://crfm.stanford.edu/helm/ewok/latest/?group=ewok_agent_properties","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM EWoK - Material Dynamics":{"url":"https://crfm.stanford.edu/helm/ewok/latest/?group=ewok_material_dynamics","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM EWoK - Material Properties":{"url":"https://crfm.stanford.edu/helm/ewok/latest/?group=ewok_material_properties","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM EWoK - Physical Dynamics":{"url":"https://crfm.stanford.edu/helm/ewok/latest/?group=ewok_physical_dynamics","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM EWoK - Physical Interactions":{"url":"https://crfm.stanford.edu/helm/ewok/latest/?group=ewok_physical_interactions","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM EWoK - Physical Relations":{"url":"https://crfm.stanford.edu/helm/ewok/latest/?group=ewok_physical_relations","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM EWoK - Social Interactions":{"url":"https://crfm.stanford.edu/helm/ewok/latest/?group=ewok_social_interactions","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM EWoK - Spatial Relations":{"url":"https://crfm.stanford.edu/helm/ewok/latest/?group=ewok_spatial_relations","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM Finance - BANKING77":{"url":"https://crfm.stanford.edu/helm/finance/latest/?group=banking77","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM Finance - FinQA":{"url":"https://crfm.stanford.edu/helm/finance/latest/?group=fin_qa","metric":"Program Accuracy","theoretical_min":0.0,"theoretical_max":100.0},"HELM Finance - FinanceBench":{"url":"https://crfm.stanford.edu/helm/finance/latest/?group=financebench","metric":"Correct Answer","theoretical_min":0.0,"theoretical_max":100.0},"HELM Lite":{"url":"https://benchmarklist.com/benchmarks/helm_lite/","metric":"Mean win rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"HELM Long Context - InfiniteBench En.MC":{"url":"https://crfm.stanford.edu/helm/long-context/latest/?group=infinite_bench_en_mc","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM Long Context - InfiniteBench En.Sum":{"url":"https://crfm.stanford.edu/helm/long-context/latest/?group=infinite_bench_en_sum","metric":"ROUGE-L","theoretical_min":0.0,"theoretical_max":100.0},"HELM Long Context - OpenAI MRCR":{"url":"https://crfm.stanford.edu/helm/long-context/latest/?group=openai_mrcr","metric":"MRCR Accuracy","theoretical_min":0.0,"theoretical_max":100.0},"HELM Long Context - RULER HotPotQA":{"url":"https://crfm.stanford.edu/helm/long-context/latest/?group=ruler_hotpotqa","metric":"RULER String Match","theoretical_min":0.0,"theoretical_max":100.0},"HELM Long Context - RULER SQuAD":{"url":"https://crfm.stanford.edu/helm/long-context/latest/?group=ruler_squad","metric":"RULER String Match","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - ACI-Bench":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=aci_bench","metric":"Jury Score","theoretical_min":0.0,"theoretical_max":null},"HELM MedHELM - ADHD-Behavior":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=shc_ptbm_med","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - ADHD-MedEffects":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=shc_sei_med","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - Administration & Workflow":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=administration_and_workflow","metric":"Mean win rate","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - CHW Care Plan":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=chw_care_plan","metric":"Jury Score","theoretical_min":0.0,"theoretical_max":5.0},"HELM MedHELM - CLEAR":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=clear","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - Clinical Decision Support":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=clinical_decision_support","metric":"Mean win rate","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - Clinical Note Generation":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=clinical_note_generation","metric":"Mean win rate","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - DischargeMe":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=dischargeme","metric":"Jury Score","theoretical_min":0.0,"theoretical_max":null},"HELM MedHELM - EHRSHOT":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=ehrshot","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - EHRSQL":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=ehr_sql","metric":"EHRSQLExeAcc","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - HeadQA":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=head_qa","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - MEDIQA":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=medi_qa","metric":"Jury Score","theoretical_min":0.0,"theoretical_max":5.0},"HELM MedHELM - MIMIC-BHC":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=mimic_bhc","metric":"Jury Score","theoretical_min":0.0,"theoretical_max":5.0},"HELM MedHELM - MIMIC-IV Billing Code":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=mimiciv_billing_code","metric":"MIMICBillingF1","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - MIMIC-RRS":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=mimic_rrs","metric":"Jury Score","theoretical_min":0.0,"theoretical_max":null},"HELM MedHELM - MTSamples":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=mtsamples_replicate","metric":"Jury Score","theoretical_min":0.0,"theoretical_max":null},"HELM MedHELM - MTSamples Procedures":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=mtsamples_procedures","metric":"Jury Score","theoretical_min":0.0,"theoretical_max":null},"HELM MedHELM - MedAlign":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=medalign","metric":"Jury Score","theoretical_min":0.0,"theoretical_max":null},"HELM MedHELM - MedCalc-Bench":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=medcalc_bench","metric":"MedCalc Accuracy","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - MedDialog":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=med_dialog","metric":"Jury Score","theoretical_min":0.0,"theoretical_max":5.0},"HELM MedHELM - MedHallu":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=medhallu","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - MedMCQA":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=med_mcqa","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - MedQA":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=med_qa","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - Medbullets":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=medbullets","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - Medec":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=medec","metric":"MedecFlagAcc","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - Medical Research":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=medical_research","metric":"Mean win rate","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - MedicationQA":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=medication_qa","metric":"Jury Score","theoretical_min":0.0,"theoretical_max":5.0},"HELM MedHELM - Mental Health":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=mental_health","metric":"Jury Score","theoretical_min":0.0,"theoretical_max":5.0},"HELM MedHELM - Patient Communication":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=patient_communication","metric":"Mean win rate","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - PubMedQA":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=pubmed_qa","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - Race-Based Medicine":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=race_based_med","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - SHC-BMT":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=shc_bmt_med","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - SHC-CDI":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=shc_cdi_med","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - SHC-Conf":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=shc_conf_med","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - SHC-ENT":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=shc_ent_med","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - SHC-GIP":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=shc_gip_med","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - SHC-Privacy":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=shc_privacy_med","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - SHC-Proxy":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=shc_proxy_med","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - SHC-Sequoia":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=shc_sequoia_med","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedHELM - STARR Patient Instructions":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=starr_patient_instructions","metric":"Jury Score","theoretical_min":0.0,"theoretical_max":5.0},"HELM MedHELM - n2c2 CT Matching":{"url":"https://crfm.stanford.edu/helm/medhelm/latest/?group=n2c2_ct_matching","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM MedQA":{"url":"https://benchmarklist.com/benchmarks/helm_medqa/","metric":"Exact match (self-reported)","theoretical_min":0.0,"theoretical_max":null},"HELM NarrativeQA":{"url":"https://crfm.stanford.edu/helm/","metric":"F1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM NaturalQuestions (Closed)":{"url":"https://crfm.stanford.edu/helm/","metric":"F1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM NaturalQuestions (Open)":{"url":"https://crfm.stanford.edu/helm/","metric":"F1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Safety":{"url":"https://benchmarklist.com/benchmarks/helm_safety/","metric":"Mean score (self-reported)","theoretical_min":0.0,"theoretical_max":null,"sources":[{"name":"HELM Safety","url":"https://benchmarklist.com/benchmarks/helm_safety/","metric":"Mean score (self-reported)"},{"name":"HELM Safety Mean score","url":"https://storage.googleapis.com/crfm-helm-public/safety/benchmark_output/releases/v1.17.0/groups/safety_scenarios.json","metric":"Score (%)"}]},"HELM Safety Anthropic Red Team":{"url":"https://storage.googleapis.com/crfm-helm-public/safety/benchmark_output/releases/v1.17.0/groups/safety_scenarios.json","metric":"LM Evaluated Safety score (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Safety BBQ":{"url":"https://storage.googleapis.com/crfm-helm-public/safety/benchmark_output/releases/v1.17.0/groups/safety_scenarios.json","metric":"BBQ accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Safety HarmBench":{"url":"https://storage.googleapis.com/crfm-helm-public/safety/benchmark_output/releases/v1.17.0/groups/safety_scenarios.json","metric":"LM Evaluated Safety score (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Safety SimpleSafetyTests":{"url":"https://storage.googleapis.com/crfm-helm-public/safety/benchmark_output/releases/v1.17.0/groups/safety_scenarios.json","metric":"LM Evaluated Safety score (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM Safety XSTest":{"url":"https://storage.googleapis.com/crfm-helm-public/safety/benchmark_output/releases/v1.17.0/groups/safety_scenarios.json","metric":"LM Evaluated Safety score (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - Flores (en-id)":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=flores_en_id","metric":"ChrF++","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - Flores (en-ta)":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=flores_en_ta","metric":"ChrF++","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - Flores (en-th)":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=flores_en_th","metric":"ChrF++","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - Flores (en-vi)":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=flores_en_vi","metric":"ChrF++","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - Flores (id-en)":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=flores_id_en","metric":"ChrF++","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - Flores (ta-en)":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=flores_ta_en","metric":"ChrF++","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - Flores (th-en)":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=flores_th_en","metric":"ChrF++","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - Flores (vi-en)":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=flores_vi_en","metric":"ChrF++","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - IndicQA":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=indicqa","metric":"SQuAD macro-averaged F1 score","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - IndicSentiment":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=indicsentiment","metric":"Macro F1 score","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - IndicXNLI":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=indicxnli","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - IndoNLI":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=indonli","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - LINDSEA Minimal Pairs (id)":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=lindsea_syntax_minimal_pairs_id","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - LINDSEA Presuppositions (id)":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=lindsea_pragmatics_presuppositions_id","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - LINDSEA Scalar Implicatures (id)":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=lindsea_pragmatics_scalar_implicatures_id","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - MLHSD":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=mlhsd","metric":"Macro F1 score","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - NusaX":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=nusax","metric":"Macro F1 score","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - Thai Toxicity Tweets":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=thaitoxicitytweets","metric":"Macro F1 score","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - TyDiQA":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=tydiqa","metric":"SQuAD macro-averaged F1 score","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - UIT-VSFC":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=uitvsfc","metric":"Macro F1 score","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - ViHSD":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=vihsd","metric":"Macro F1 score","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - Wisesight":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=wisesight","metric":"Macro F1 score","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - XCOPA (id)":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=xcopa_id","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - XCOPA (ta)":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=xcopa_ta","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - XCOPA (th)":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=xcopa_th","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - XCOPA (vi)":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=xcopa_vi","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - XNLI (th)":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=xnli_th","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - XNLI (vi)":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=xnli_vi","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - XQuAD (Thai)":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=xquad_th","metric":"SQuAD macro-averaged F1 score","theoretical_min":0.0,"theoretical_max":100.0},"HELM SeaHELM - XQuAD (Vietnamese)":{"url":"https://crfm.stanford.edu/helm/seahelm/latest/?group=xquad_vi","metric":"SQuAD macro-averaged F1 score","theoretical_min":0.0,"theoretical_max":100.0},"HELM ThaiExam - A-Level":{"url":"https://crfm.stanford.edu/helm/thaiexam/latest/?group=thai_exam_a_level","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM ThaiExam - IC":{"url":"https://crfm.stanford.edu/helm/thaiexam/latest/?group=thai_exam_ic","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM ThaiExam - ONET":{"url":"https://crfm.stanford.edu/helm/thaiexam/latest/?group=thai_exam_onet","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM ThaiExam - TGAT":{"url":"https://crfm.stanford.edu/helm/thaiexam/latest/?group=thai_exam_tgat","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM ThaiExam - TPAT-1":{"url":"https://crfm.stanford.edu/helm/thaiexam/latest/?group=thai_exam_tpat1","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM ThaiExam - ThaiExam":{"url":"https://crfm.stanford.edu/helm/thaiexam/latest/?group=thai_exam","metric":"EM","theoretical_min":0.0,"theoretical_max":100.0},"HELM WMT 2014":{"url":"https://crfm.stanford.edu/helm/","metric":"BLEU-4 (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM v2 Lite - HumanEval (Code)":{"url":"https://nlp.stanford.edu/helm/v2-lite-finch/?group=code_humaneval","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM v2 Lite - LegalBench":{"url":"https://nlp.stanford.edu/helm/v2-lite-finch/?group=legalbench","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELM v2 Lite - MedQA":{"url":"https://nlp.stanford.edu/helm/v2-lite-finch/?group=med_qa","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"HELMET (128K)":{"url":"https://princeton-nlp.github.io/HELMET","metric":"Average Score","theoretical_min":0.0,"theoretical_max":100.0},"HIDBench":{"url":"https://benchmarklist.com/benchmarks/hidbench/","metric":"DARPA E3 CADETS MCC (self-reported)","theoretical_min":0.0,"theoretical_max":null},"HLCE":{"url":"https://github.com/Humanity-s-Last-Code-Exam/HLCE","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"HMMT February 2026":{"url":"https://benchmarklist.com/benchmarks/hmmt_feb_2026/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"HOLMES":{"url":"https://benchmarklist.com/benchmarks/holmes_evaluating_higher_order_logical_reasoning_in_llms/","metric":"Overall Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"HREF":{"url":"https://huggingface.co/spaces/allenai/href","metric":"Average HREF Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"HSCodeComp - 10-digit":{"url":"https://arxiv.org/html/2510.19631v1","metric":"Exact Match Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HSCodeComp - 10-digit (Vision)":{"url":"https://arxiv.org/html/2510.19631v1","metric":"Exact Match Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HSCodeComp - 2-digit":{"url":"https://arxiv.org/html/2510.19631v1","metric":"Exact Match Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HSCodeComp - 2-digit (Vision)":{"url":"https://arxiv.org/html/2510.19631v1","metric":"Exact Match Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HSCodeComp - 4-digit":{"url":"https://arxiv.org/html/2510.19631v1","metric":"Exact Match Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HSCodeComp - 4-digit (Vision)":{"url":"https://arxiv.org/html/2510.19631v1","metric":"Exact Match Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HSCodeComp - 6-digit":{"url":"https://arxiv.org/html/2510.19631v1","metric":"Exact Match Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HSCodeComp - 6-digit (Vision)":{"url":"https://arxiv.org/html/2510.19631v1","metric":"Exact Match Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HSCodeComp - 8-digit":{"url":"https://arxiv.org/html/2510.19631v1","metric":"Exact Match Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HSCodeComp - 8-digit (Vision)":{"url":"https://arxiv.org/html/2510.19631v1","metric":"Exact Match Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Hack-Verifiable TextArena":{"url":"https://benchmarklist.com/benchmarks/hack_verifiable_textarena/","metric":"Avg HR (self-reported)","theoretical_min":0.0,"theoretical_max":null},"HalluHard":{"url":"https://halluhard.com/","metric":"Turn-1 Hallucination Rate","theoretical_min":0.0,"theoretical_max":1.0},"HalluScore":{"url":"https://benchmarklist.com/benchmarks/halluscore/","metric":"Faith% (self-reported)","theoretical_min":0.0,"theoretical_max":null},"HalluWorld":{"url":"https://benchmarklist.com/benchmarks/halluworld_a_controlled_benchmark_for_hallucination_via_reference_world_models/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"HallusionBench":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Overall Score","theoretical_min":0.0,"theoretical_max":100.0,"sources":[{"name":"HallusionBench","url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Overall Score"},{"name":"OpenCompass Multimodal - HallusionBench","url":"https://rank.opencompass.org.cn","metric":"Score (%)"}]},"Halluverse-M3":{"url":"https://benchmarklist.com/benchmarks/halluverse_m3/","metric":"Macro Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"HardMTBench":{"url":"https://benchmarklist.com/benchmarks/hardmtbench_stress_testing_chinese_english_translation_on_knowledge_intensive_domains/","metric":"HardMTBench zh-en GEMBA-DA (self-reported)","theoretical_min":0.0,"theoretical_max":null},"HardcoreLogic - Increased Complexity":{"url":"https://huggingface.co/spaces/JunsWan/HardcoreLogic","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HardcoreLogic - Total":{"url":"https://huggingface.co/spaces/JunsWan/HardcoreLogic","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HardcoreLogic - Uncommon Elements":{"url":"https://huggingface.co/spaces/JunsWan/HardcoreLogic","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HardcoreLogic - Unsolvable Puzzles":{"url":"https://huggingface.co/spaces/JunsWan/HardcoreLogic","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HarmActionsEval":{"url":"https://benchmarklist.com/benchmarks/harmactions_eval/","metric":"SafeActions@1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"HarmVideoBench":{"url":"https://benchmarklist.com/benchmarks/harmvideobench_benchmarking_harmful_video_understanding_in_large_multimodal_models/","metric":"Macro Avg. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Harvey Legal Agent Benchmark":{"url":"https://benchmarklist.com/benchmarks/harvey_lab/","metric":"All-Pass Task Success (self-reported)","theoretical_min":0.0,"theoretical_max":null},"HealthAdminBench":{"url":"https://leaderboard.steel.dev/leaderboards/healthadminbench/","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"HealthBench":{"url":"https://openai.com/index/healthbench/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"HealthBench (Fable/Mythos)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"HealthBench Hard":{"url":"https://benchmarklist.com/benchmarks/healthbench_hard/","metric":"Overall score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"HealthBench Professional (Fable/Mythos)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Hebrew LLM - Nikud":{"url":"https://huggingface.co/spaces/hebrew-llm-leaderboard/leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Hebrew LLM - Summarization (Pairwise)":{"url":"https://huggingface.co/spaces/hebrew-llm-leaderboard/leaderboard","metric":"Pairwise Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Hebrew LLM - Translation (Pairwise)":{"url":"https://huggingface.co/spaces/hebrew-llm-leaderboard/leaderboard","metric":"Pairwise Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Hebrew LLM - Trivia (0-shot)":{"url":"https://huggingface.co/spaces/hebrew-llm-leaderboard/leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Hebrew LLM - Winograd (0-shot)":{"url":"https://huggingface.co/spaces/hebrew-llm-leaderboard/leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HellaSwag":{"url":"https://epoch.ai/benchmarks/eci","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":25},"HiL-Bench":{"url":"https://labs.scale.com/leaderboard/hil","metric":"Combined Pass@3 (%)","theoretical_min":0.0,"theoretical_max":100.0},"HieroglyphBench":{"url":"https://boggs.tech/posts/reading-hieroglyphs-with-vlms/","metric":"Sign accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"HindiGen":{"url":"https://huggingface.co/spaces/inceptionai/Leaderboards","metric":"3C3H Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Honeyval":{"url":"https://benchmarklist.com/benchmarks/honeyval_a_comprehensive_evaluation_framework_for_llm_powered_http_honeypots/","metric":"Interaction Length (self-reported)","theoretical_min":0.0,"theoretical_max":null},"HoudiniVexBench":{"url":"https://huggingface.co/datasets/kelvincai/HoudiniVexBench","metric":"Overall Score","theoretical_min":0.0,"theoretical_max":1.0},"HumainE Leaderboard":{"url":"https://huggingface.co/spaces/ProlificAI/humaine-leaderboard","metric":"HumainE Score","theoretical_min":0.0,"theoretical_max":100.0},"Human-Grounded Multimodal Benchmark with 900K":{"url":"https://benchmarklist.com/benchmarks/human_grounded_multimodal_benchmark_with_900k_scale_aggregated_student_response_distributions_from_japan_s_national_assessment_of_academic_a/","metric":"Science MC Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"HumanEval+":{"url":"https://benchmarklist.com/benchmarks/humaneval/","metric":"HumanEval+ pass@1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"HumanLikeness - Discourse-1":{"url":"https://huggingface.co/spaces/xufengduan/HumanLikeness","metric":"Humanlike Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"HumanLikeness - Discourse-2":{"url":"https://huggingface.co/spaces/xufengduan/HumanLikeness","metric":"Humanlike Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"HumanLikeness - Meaning-1":{"url":"https://huggingface.co/spaces/xufengduan/HumanLikeness","metric":"Humanlike Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"HumanLikeness - Meaning-2":{"url":"https://huggingface.co/spaces/xufengduan/HumanLikeness","metric":"Humanlike Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"HumanLikeness - Overall":{"url":"https://huggingface.co/spaces/xufengduan/HumanLikeness","metric":"Overall Humanlike (%)","theoretical_min":0.0,"theoretical_max":100.0},"HumanLikeness - Sound-1":{"url":"https://huggingface.co/spaces/xufengduan/HumanLikeness","metric":"Humanlike Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"HumanLikeness - Sound-2":{"url":"https://huggingface.co/spaces/xufengduan/HumanLikeness","metric":"Humanlike Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"HumanLikeness - Syntax-1":{"url":"https://huggingface.co/spaces/xufengduan/HumanLikeness","metric":"Humanlike Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"HumanLikeness - Syntax-2":{"url":"https://huggingface.co/spaces/xufengduan/HumanLikeness","metric":"Humanlike Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"HumanLikeness - Word-1":{"url":"https://huggingface.co/spaces/xufengduan/HumanLikeness","metric":"Humanlike Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"HumanLikeness - Word-2":{"url":"https://huggingface.co/spaces/xufengduan/HumanLikeness","metric":"Humanlike Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Humanity's Last Exam (Anthropic No Tools)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Humanity's Last Exam (Anthropic Tools)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Humanity's Last Exam (Fable/Mythos No Tools)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Humanity's Last Exam (Fable/Mythos Tools)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"IDE-Bench":{"url":"https://benchmarklist.com/benchmarks/ide_bench/","metric":"Pass@1 Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"IDP Leaderboard":{"url":"https://idp-leaderboard.org/","metric":"Avg Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"IFEval Leaderboard":{"url":"https://huggingface.co/spaces/Krisseck/IFEval-Leaderboard","metric":"Final Score","theoretical_min":0.0,"theoretical_max":100.0},"IFMTBench":{"url":"https://benchmarklist.com/benchmarks/ifmtbench_a_comprehensive_benchmark_for_multilingual_translation_instruction_following/","metric":"IF$_\\text{T}$ (self-reported)","theoretical_min":0.0,"theoretical_max":null},"IMO-AnswerBench":{"url":"https://benchmarklist.com/benchmarks/imo_answerbench/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"IMO-Bench":{"url":"https://imobench.github.io/","metric":"Advanced ProofBench Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"INCLUDE":{"url":"https://benchmarklist.com/benchmarks/include_benchmark/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"INCLUDE-base-44 European Languages":{"url":"https://huggingface.co/spaces/speakleash/include-base-european-leaderboard","metric":"Average Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"INTIMA Companionship":{"url":"https://huggingface.co/spaces/frimelle/companionship-leaderboard","metric":"Companionship score (%)","theoretical_min":0.0,"theoretical_max":100.0},"INVESTORBENCH":{"url":"https://benchmarklist.com/benchmarks/investorbench/","metric":"Average stock cumulative return (self-reported)","theoretical_min":null,"theoretical_max":null},"IOI":{"url":"https://benchmarklist.com/benchmarks/vals_ioi/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ITBench-AA":{"url":"https://benchmarklist.com/benchmarks/itbench_aa/","metric":"Average Precision at Full Recall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"IUMB":{"url":"https://pellaml.github.io/iumb/#benchmark","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Icelandic LLM - ARC-Challenge-IS":{"url":"https://huggingface.co/spaces/mideind/icelandic-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Icelandic LLM - Belebele-IS":{"url":"https://huggingface.co/spaces/mideind/icelandic-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Icelandic LLM - GED":{"url":"https://huggingface.co/spaces/mideind/icelandic-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Icelandic LLM - Inflection":{"url":"https://huggingface.co/spaces/mideind/icelandic-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Icelandic LLM - WikiQA-IS":{"url":"https://huggingface.co/spaces/mideind/icelandic-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Icelandic LLM - WinoGrande-IS":{"url":"https://huggingface.co/spaces/mideind/icelandic-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Icelandic LLM Leaderboard - Average":{"url":"https://huggingface.co/spaces/mideind/icelandic-llm-leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"IgakuQA119":{"url":"https://github.com/naoto-iwase/IgakuQA119","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"IllusionVQA":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"Accuracy","theoretical_min":0.0,"theoretical_max":100.0},"ImplicitQA (Average Accuracy)":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"Average Accuracy","theoretical_min":0.0,"theoretical_max":100.0},"ImplicitQA (Macro Average Accuracy)":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"Macro Average Accuracy","theoretical_min":0.0,"theoretical_max":100.0},"IndoBias":{"url":"https://benchmarklist.com/benchmarks/indobias_a_dual_track_culturally_grounded_benchmark_for_llms_bias_evaluation_in_indonesian_languages/","metric":"Ideology and Religion IND (self-reported)","theoretical_min":0.0,"theoretical_max":null},"IndustryBench":{"url":"https://benchmarklist.com/benchmarks/industrybench_probing_the_industrial_knowledge_boundaries_of_llms/","metric":"Final (SV) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"IneqMath":{"url":"https://benchmarklist.com/benchmarks/ineqmath/","metric":"Overall Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"InferenceBench":{"url":"https://inferencebench.ai/","metric":"Speedup Score","theoretical_min":0.0,"theoretical_max":null},"InfiBench":{"url":"https://infi-coder.github.io/infibench/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"InfiniteBench":{"url":"https://github.com/OpenBMB/InfiniteBench","metric":"Avg Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"InsanityBench":{"url":"https://robinhaselhorst.com/","metric":"Total Score","theoretical_min":0.0,"theoretical_max":100.0},"IntentGrasp":{"url":"https://benchmarklist.com/benchmarks/intentgrasp/","metric":"Overall (All Set) Avg (Std) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"InterCode":{"url":"https://intercode-benchmark.github.io","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"InterLV-Search":{"url":"https://benchmarklist.com/benchmarks/interlv_search/","metric":"Level 3 +Tool Avg (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Inverse Turing Bench":{"url":"https://benchmarklist.com/benchmarks/inverse_turing_bench/","metric":"Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"InvisibleBench":{"url":"https://benchmarklist.com/benchmarks/invisiblebench/","metric":"Hard Fail Rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"IslamicLegalBench":{"url":"https://huggingface.co/spaces/AbdullahMushtaq/IslamicLegalBench-Leaderboard","metric":"Weighted Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"IslamicLegalBench - High Complexity":{"url":"https://huggingface.co/spaces/AbdullahMushtaq/IslamicLegalBench-Leaderboard","metric":"Weighted Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"IslamicLegalBench - Low Complexity":{"url":"https://huggingface.co/spaces/AbdullahMushtaq/IslamicLegalBench-Leaderboard","metric":"Weighted Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"IslamicLegalBench - Moderate Complexity":{"url":"https://huggingface.co/spaces/AbdullahMushtaq/IslamicLegalBench-Leaderboard","metric":"Weighted Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"J1-ENVS":{"url":"https://benchmarklist.com/benchmarks/j1_envs/","metric":"Overall Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"JEE/NEET LLM Benchmark - JEE Advanced 2025":{"url":"https://huggingface.co/datasets/Reja1/jee-neet-benchmark","metric":"Exam Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"JEE/NEET LLM Benchmark - JEE Advanced 2026":{"url":"https://huggingface.co/datasets/Reja1/jee-neet-benchmark","metric":"Exam Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"JEE/NEET LLM Benchmark - NEET 2025":{"url":"https://huggingface.co/datasets/Reja1/jee-neet-benchmark","metric":"Exam Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"JEE/NEET LLM Benchmark - NEET 2026":{"url":"https://huggingface.co/datasets/Reja1/jee-neet-benchmark","metric":"Exam Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"JMMMU-Pro - Culture Agnostic":{"url":"https://huggingface.co/spaces/JMMMU/JMMMU-Pro_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"JMMMU-Pro - Culture Specific":{"url":"https://huggingface.co/spaces/JMMMU/JMMMU-Pro_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"JMMMU-Pro - Japanese Art":{"url":"https://huggingface.co/spaces/JMMMU/JMMMU-Pro_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"JMMMU-Pro - Japanese Heritage":{"url":"https://huggingface.co/spaces/JMMMU/JMMMU-Pro_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"JMMMU-Pro - Japanese History":{"url":"https://huggingface.co/spaces/JMMMU/JMMMU-Pro_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"JMMMU-Pro - Overall":{"url":"https://huggingface.co/spaces/JMMMU/JMMMU-Pro_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"JMMMU-Pro - World History":{"url":"https://huggingface.co/spaces/JMMMU/JMMMU-Pro_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"JMed48k":{"url":"https://benchmarklist.com/benchmarks/jmed48k_a_multi_profession_japanese_medical_licensing_benchmark_for_vision_language_model_evaluation/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"JamSet/JamBench":{"url":"https://benchmarklist.com/benchmarks/jamset_jambench/","metric":"Task 1a SCS (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Journalistic Bias Accuracy":{"url":"https://github.com/wandering-isle/journalistic-bias","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Journalistic Bias F1-macro":{"url":"https://github.com/wandering-isle/journalistic-bias","metric":"F1-Macro (%)","theoretical_min":0.0,"theoretical_max":100.0},"JuICE":{"url":"https://benchmarklist.com/benchmarks/juice_a_benchmark_for_evaluating_llm_judge_in_identifying_cultural_errors/","metric":"F1 Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Judge Arena":{"url":"https://huggingface.co/spaces/AtlaAI/judge-arena","metric":"ELO Score","theoretical_min":null,"theoretical_max":null},"JudgeBench Coding":{"url":"https://huggingface.co/datasets/ScalerLab/JudgeBench","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"JudgeBench Knowledge":{"url":"https://huggingface.co/datasets/ScalerLab/JudgeBench","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"JudgeBench Math":{"url":"https://huggingface.co/datasets/ScalerLab/JudgeBench","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"JudgeBench Reasoning":{"url":"https://huggingface.co/datasets/ScalerLab/JudgeBench","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Judgemark v2.1":{"url":"https://eqbench.com/judgemark-v2.html","metric":"Judgemark Score (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"JustEval":{"url":"https://allenai.github.io/re-align/just_eval.html","metric":"Avg Score (1-5)","theoretical_min":0.0,"theoretical_max":5.0},"JustEval - Clarity":{"url":"https://allenai.github.io/re-align/just_eval.html","metric":"Score (1-5)","theoretical_min":0.0,"theoretical_max":5.0},"JustEval - Depth":{"url":"https://allenai.github.io/re-align/just_eval.html","metric":"Score (1-5)","theoretical_min":0.0,"theoretical_max":5.0},"JustEval - Engagement":{"url":"https://allenai.github.io/re-align/just_eval.html","metric":"Score (1-5)","theoretical_min":0.0,"theoretical_max":5.0},"JustEval - Factuality":{"url":"https://allenai.github.io/re-align/just_eval.html","metric":"Score (1-5)","theoretical_min":0.0,"theoretical_max":5.0},"JustEval - Helpfulness":{"url":"https://allenai.github.io/re-align/just_eval.html","metric":"Score (1-5)","theoretical_min":0.0,"theoretical_max":5.0},"JustEval - Safety":{"url":"https://allenai.github.io/re-align/just_eval.html","metric":"Score (1-5)","theoretical_min":0.0,"theoretical_max":5.0},"K-12EduBench":{"url":"https://benchmarklist.com/benchmarks/k12edubench/","metric":"Lambda_js Avg (self-reported)","theoretical_min":0.0,"theoretical_max":null},"K-FinHallu":{"url":"https://benchmarklist.com/benchmarks/k_finhallu_a_hallucination_detection_benchmark_for_multi_turn_rag_in_korean_finance/","metric":"Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"K-MetBench":{"url":"https://benchmarklist.com/benchmarks/kmetbench/","metric":"Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"K12-Bench":{"url":"https://benchmarklist.com/benchmarks/k12_bench/","metric":"Overall EM (self-reported)","theoretical_min":0.0,"theoretical_max":null},"KINA":{"url":"https://arxiv.org/html/2606.05104v2","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"KOFFVQA":{"url":"https://huggingface.co/spaces/maum-ai/KOFFVQA-Leaderboard","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"KOFFVQA - Commonsense Reasoning":{"url":"https://huggingface.co/spaces/maum-ai/KOFFVQA-Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"KOFFVQA - Document Understanding":{"url":"https://huggingface.co/spaces/maum-ai/KOFFVQA-Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"KOFFVQA - Graph and Chart Understanding":{"url":"https://huggingface.co/spaces/maum-ai/KOFFVQA-Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"KOFFVQA - Hallucination and Robustness":{"url":"https://huggingface.co/spaces/maum-ai/KOFFVQA-Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"KOFFVQA - Korean OCR":{"url":"https://huggingface.co/spaces/maum-ai/KOFFVQA-Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"KOFFVQA - Korean Recognition":{"url":"https://huggingface.co/spaces/maum-ai/KOFFVQA-Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"KOFFVQA - Object Attributes":{"url":"https://huggingface.co/spaces/maum-ai/KOFFVQA-Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"KOFFVQA - Recognition":{"url":"https://huggingface.co/spaces/maum-ai/KOFFVQA-Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"KOFFVQA - Relationship":{"url":"https://huggingface.co/spaces/maum-ai/KOFFVQA-Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"KOFFVQA - Table Understanding":{"url":"https://huggingface.co/spaces/maum-ai/KOFFVQA-Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"KORGym":{"url":"https://razor233.github.io/KORGYM_HomePage/","metric":"Score","theoretical_min":0.0,"theoretical_max":1.0},"KORGym - Control and Interaction":{"url":"https://razor233.github.io/KORGYM_HomePage/","metric":"Score","theoretical_min":0.0,"theoretical_max":1.0},"KORGym - Mathematical and Logical":{"url":"https://razor233.github.io/KORGYM_HomePage/","metric":"Score","theoretical_min":0.0,"theoretical_max":1.0},"KORGym - Puzzle":{"url":"https://razor233.github.io/KORGYM_HomePage/","metric":"Score","theoretical_min":0.0,"theoretical_max":1.0},"KORGym - Spatial and Geometric":{"url":"https://razor233.github.io/KORGYM_HomePage/","metric":"Score","theoretical_min":0.0,"theoretical_max":1.0},"KORGym - Strategic":{"url":"https://razor233.github.io/KORGYM_HomePage/","metric":"Score","theoretical_min":0.0,"theoretical_max":1.0},"Kaggle DeepSearchQA (Google)":{"url":"https://www.kaggle.com/benchmarks/google/dsqa","metric":"F1 Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Kaggle Enterprise Ops (IBM)":{"url":"https://www.kaggle.com/benchmarks/ibm-research/enterprise-ops","metric":"Avg Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Kaggle FACTS (Google)":{"url":"https://www.kaggle.com/benchmarks/google/facts","metric":"Avg Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Kaggle FACTS Grounding":{"url":"https://www.kaggle.com/benchmarks/google/facts-grounding/versions/2","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Kaggle FACTS Multimodal":{"url":"https://www.kaggle.com/benchmarks/google/facts-multimodal/versions/1","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Kaggle FACTS Parametric":{"url":"https://www.kaggle.com/benchmarks/google/facts-parametric/versions/1","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Kaggle FACTS Search":{"url":"https://www.kaggle.com/benchmarks/google/facts-search/versions/1","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Kaggle Game Arena Chess":{"url":"https://www.kaggle.com/benchmarks/kaggle/chess","metric":"Elo Rating","theoretical_min":null,"theoretical_max":null},"Kaggle Game Arena Chess (Openings)":{"url":"https://www.kaggle.com/benchmarks/kaggle/chess","metric":"Elo Rating","theoretical_min":null,"theoretical_max":null},"Kaggle Game Arena Four in a Row":{"url":"https://www.kaggle.com/benchmarks/kaggle/four-in-a-row","metric":"Elo Rating","theoretical_min":null,"theoretical_max":null},"Kaggle Game Arena Poker (Heads Up)":{"url":"https://www.kaggle.com/benchmarks/kaggle/poker-heads-up","metric":"Mean BB/100","theoretical_min":null,"theoretical_max":null},"Kaggle Game Arena Werewolf":{"url":"https://www.kaggle.com/benchmarks/kaggle/werewolf","metric":"Equilibrium Rating","theoretical_min":null,"theoretical_max":null},"Kaggle WWTP Engineering":{"url":"https://www.kaggle.com/benchmarks/mehmetisik/wwtp-engineering-benchmark","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Kagi LLM Benchmark":{"url":"https://help.kagi.com/kagi/ai/llm-benchmark.html","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"KataGo-Bench-1K":{"url":"https://arxiv.org/abs/2601.16447","metric":"Next-move Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Kazakh LLM - Biology UNT":{"url":"https://huggingface.co/spaces/kz-transformers/kaz-llm-lb","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"sources":[{"name":"Kazakh LLM - Biology UNT","url":"https://huggingface.co/spaces/kz-transformers/kaz-llm-lb","metric":"Accuracy (%)"},{"name":"Kaz-LLM - Biology (UNT)","url":"https://huggingface.co/spaces/kz-transformers/kaz-llm-lb","metric":"Accuracy (%)"}]},"Kazakh LLM - Constitution":{"url":"https://huggingface.co/spaces/kz-transformers/kaz-llm-lb","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Kazakh LLM - Dastur":{"url":"https://huggingface.co/spaces/kz-transformers/kaz-llm-lb","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Kazakh LLM - English":{"url":"https://huggingface.co/spaces/kz-transformers/kaz-llm-lb","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Kazakh LLM - Geography":{"url":"https://huggingface.co/spaces/kz-transformers/kaz-llm-lb","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"sources":[{"name":"Kazakh LLM - Geography","url":"https://huggingface.co/spaces/kz-transformers/kaz-llm-lb","metric":"Accuracy (%)"},{"name":"Kaz-LLM - Geography (UNT)","url":"https://huggingface.co/spaces/kz-transformers/kaz-llm-lb","metric":"Accuracy (%)"}]},"Kazakh LLM - History of Kazakhstan":{"url":"https://huggingface.co/spaces/kz-transformers/kaz-llm-lb","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"sources":[{"name":"Kazakh LLM - History of Kazakhstan","url":"https://huggingface.co/spaces/kz-transformers/kaz-llm-lb","metric":"Accuracy (%)"},{"name":"Kaz-LLM - History of Kazakhstan (UNT)","url":"https://huggingface.co/spaces/kz-transformers/kaz-llm-lb","metric":"Accuracy (%)"}]},"Kazakh LLM - Human Society & Rights":{"url":"https://huggingface.co/spaces/kz-transformers/kaz-llm-lb","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Kazakh LLM - Kazakh Language & Literature":{"url":"https://huggingface.co/spaces/kz-transformers/kaz-llm-lb","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"sources":[{"name":"Kazakh LLM - Kazakh Language & Literature","url":"https://huggingface.co/spaces/kz-transformers/kaz-llm-lb","metric":"Accuracy (%)"},{"name":"Kaz-LLM - Kazakh Language & Literature (UNT)","url":"https://huggingface.co/spaces/kz-transformers/kaz-llm-lb","metric":"Accuracy (%)"}]},"Kazakh LLM - MMLU (translated)":{"url":"https://huggingface.co/spaces/kz-transformers/kaz-llm-lb","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"sources":[{"name":"Kazakh LLM - MMLU (translated)","url":"https://huggingface.co/spaces/kz-transformers/kaz-llm-lb","metric":"Accuracy (%)"},{"name":"Kaz-LLM - MMLU (Kazakh)","url":"https://huggingface.co/spaces/kz-transformers/kaz-llm-lb","metric":"Accuracy (%)"}]},"Kazakh LLM - World History":{"url":"https://huggingface.co/spaces/kz-transformers/kaz-llm-lb","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Kernel Arena - KernelBench HIP":{"url":"https://kernelarena.ai/eval?suite=kernelbench-hip","metric":"Mean Correctness+Speedup","theoretical_min":0.0,"theoretical_max":null},"Kernel Arena - WaferBench NVFP4":{"url":"https://kernelarena.ai/eval?suite=waferbench-nvfp4","metric":"Mean Correctness+Speedup","theoretical_min":0.0,"theoretical_max":null},"KernelBench":{"url":"https://scalingintelligence.stanford.edu/KernelBenchLeaderboard/","metric":"Avg Speedup vs PyTorch","theoretical_min":null,"theoretical_max":null},"KernelBench Hub - Mega":{"url":"https://kernelbench.com/mega","metric":"Best Speedup vs Reference (x)","theoretical_min":0.0,"theoretical_max":null},"Klu LLM Leaderboard":{"url":"https://klu.ai/llm-leaderboard","metric":"Klu Index","theoretical_min":0.0,"theoretical_max":100.0},"Kluster Hallucination Detection - Non-RAG Resistance":{"url":"https://huggingface.co/spaces/kluster-ai/LLM-Hallucination-Detection-Leaderboard","metric":"Resistance (100 - Hallucination Rate %)","theoretical_min":0.0,"theoretical_max":100.0},"Kluster Hallucination Detection - RAG Method 1 Resistance":{"url":"https://huggingface.co/spaces/kluster-ai/LLM-Hallucination-Detection-Leaderboard","metric":"Resistance (100 - Hallucination Rate %)","theoretical_min":0.0,"theoretical_max":100.0},"Kluster Hallucination Detection - RAG Method 2 Resistance":{"url":"https://huggingface.co/spaces/kluster-ai/LLM-Hallucination-Detection-Leaderboard","metric":"Resistance (100 - Hallucination Rate %)","theoretical_min":0.0,"theoretical_max":100.0},"Kluster Hallucination Detection - RAG Method 3 Resistance":{"url":"https://huggingface.co/spaces/kluster-ai/LLM-Hallucination-Detection-Leaderboard","metric":"Resistance (100 - Hallucination Rate %)","theoretical_min":0.0,"theoretical_max":100.0},"Kluster Hallucination Detection - RAG Resistance":{"url":"https://huggingface.co/spaces/kluster-ai/LLM-Hallucination-Detection-Leaderboard","metric":"Resistance (100 - Hallucination Rate %)","theoretical_min":0.0,"theoretical_max":100.0},"KnotBench":{"url":"https://benchmarklist.com/benchmarks/knotbench/","metric":"Accuracy (%) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Ko-AgentBench - L1 Single Tool Call":{"url":"https://huggingface.co/spaces/huggingface-KREW/Ko-AgentBench","metric":"Argument Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ko-AgentBench - L3 Sequential Tool Reasoning":{"url":"https://huggingface.co/spaces/huggingface-KREW/Ko-AgentBench","metric":"PSM (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ko-AgentBench - L4 Parallel Tool Reasoning":{"url":"https://huggingface.co/spaces/huggingface-KREW/Ko-AgentBench","metric":"Coverage (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ko-AgentBench - L5 Error Handling & Robustness":{"url":"https://huggingface.co/spaces/huggingface-KREW/Ko-AgentBench","metric":"Adaptive Routing Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ko-AgentBench - L6 Efficient Tool Utilization":{"url":"https://huggingface.co/spaces/huggingface-KREW/Ko-AgentBench","metric":"Efficiency Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ko-AgentBench - L7 Long-Context Memory":{"url":"https://huggingface.co/spaces/huggingface-KREW/Ko-AgentBench","metric":"Context Retention (%)","theoretical_min":0.0,"theoretical_max":100.0},"Konkur 1404 - Art":{"url":"https://huggingface.co/spaces/mshojaei77/konkur1404-llm-leaderboard","metric":"Accuracy (%, text-only)","theoretical_min":0.0,"theoretical_max":100.0},"Konkur 1404 - Experimental Sciences":{"url":"https://huggingface.co/spaces/mshojaei77/konkur1404-llm-leaderboard","metric":"Accuracy (%, text-only)","theoretical_min":0.0,"theoretical_max":100.0},"Konkur 1404 - Foreign Language":{"url":"https://huggingface.co/spaces/mshojaei77/konkur1404-llm-leaderboard","metric":"Accuracy (%, text-only)","theoretical_min":0.0,"theoretical_max":100.0},"Konkur 1404 - Humanities":{"url":"https://huggingface.co/spaces/mshojaei77/konkur1404-llm-leaderboard","metric":"Accuracy (%, text-only)","theoretical_min":0.0,"theoretical_max":100.0},"Konkur 1404 - Mathematics":{"url":"https://huggingface.co/spaces/mshojaei77/konkur1404-llm-leaderboard","metric":"Accuracy (%, text-only)","theoretical_min":0.0,"theoretical_max":100.0},"Konkur 1404 - Overall":{"url":"https://huggingface.co/spaces/mshojaei77/konkur1404-llm-leaderboard","metric":"Accuracy (%, text-only)","theoretical_min":0.0,"theoretical_max":100.0},"Korean Canonical Legal Benchmark":{"url":"https://benchmarklist.com/benchmarks/kcl_legal_benchmark/","metric":"Mean with supporting precedents (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Kotlin-bench":{"url":"https://firebender.com/leaderboard","metric":"Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"LA Leaderboard":{"url":"https://huggingface.co/spaces/la-leaderboard/la-leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LA Leaderboard - AQuAS":{"url":"https://huggingface.co/spaces/la-leaderboard/la-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LA Leaderboard - COPA Spanish":{"url":"https://huggingface.co/spaces/la-leaderboard/la-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LA Leaderboard - EusExams Basque":{"url":"https://huggingface.co/spaces/la-leaderboard/la-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LA Leaderboard - GalCoLA":{"url":"https://huggingface.co/spaces/la-leaderboard/la-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LA Leaderboard - PIQA Catalan":{"url":"https://huggingface.co/spaces/la-leaderboard/la-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LA Leaderboard - Spanish Law Exams":{"url":"https://huggingface.co/spaces/la-leaderboard/la-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LA Leaderboard - XNLI Galician":{"url":"https://huggingface.co/spaces/la-leaderboard/la-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LAB-Bench FigQA (Anthropic No Tools)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LAB-Bench FigQA (Anthropic Tools)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LAB-Bench FigQA (Mythos Preview No Tools)":{"url":"https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LAB-Bench FigQA (Mythos Preview Tools)":{"url":"https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LAB-Bench FigQA No Tools (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LAB-Bench FigQA With Cropping (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LABBench2 - Clinical Trial Questions (Anthropic)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LABBench2 - Patent Questions (Anthropic)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LABBench2 - Supplementary Materials (Anthropic)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LABBench2 - Table Reading (Anthropic)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LABBench2 Clinical Trials":{"url":"https://benchmarklist.com/benchmarks/labbench2_clinical_trials/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LABBench2 Patent Questions":{"url":"https://benchmarklist.com/benchmarks/labbench2_patents/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LEXam":{"url":"https://lexam-benchmark.github.io/","metric":"Multiple-Choice Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - LibrusecHistory":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - LibrusecMHQA":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - LibrusecMHQA *":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - LongContextMultiQ":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - LongContextMultiQ *":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - MatreshkaNames":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - MatreshkaNames *":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - MatreshkaYesNo":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - Passkey":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - PasskeyWithLibrusec":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - ru2WikiMultihopQA":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - ru2WikiMultihopQA *":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - ruBABILongQA1":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - ruBABILongQA2":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - ruBABILongQA3":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - ruBABILongQA3 *":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - ruBABILongQA4":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - ruBABILongQA5":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - ruGSM100":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - ruQasper":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - ruQuALITY":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - ruSciAbstractRetrieval":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - ruSciFi":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - ruSciPassageCount":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - ruSciPassageCount *":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - ruTPO":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LIBRA - ruTREC":{"url":"https://huggingface.co/spaces/ai-forever/LIBRA-Leaderboard","metric":"Dataset Total Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LITMUS":{"url":"https://benchmarklist.com/benchmarks/litmus_benchmarking_behavioral_jailbreaks_of_llm_agents_in_real_os_environments/","metric":"Attack Success Rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LLM Arena RU":{"url":"https://llmarena.ru/","metric":"Arena Elo","theoretical_min":null,"theoretical_max":null},"LLM Benchmarker Suite":{"url":"https://github.com/FormulaMonks/llm-benchmarker-suite","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Chess (Saplin)":{"url":"https://maxim-saplin.github.io/llm_chess/","metric":"ELO","theoretical_min":null,"theoretical_max":null},"LLM Emergent Collusion":{"url":"https://github.com/lechmazur/emergent_collusion","metric":"Collusion Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM GPU Kernel Generation":{"url":"https://arxiv.org/abs/2607.14541","metric":"Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Game Benchmark":{"url":"https://benchmarklist.com/benchmarks/llm_game_benchmark/","metric":"Source Win Rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LLM Grid Game Benchmark":{"url":"https://research-outcome.github.io/LLM-Game-Benchmark/leaderboard/","metric":"Avg Win Rate vs Random (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Hallucination Detection - Non-RAG":{"url":"https://huggingface.co/spaces/kluster-ai/LLM-Hallucination-Detection-Leaderboard","metric":"Hallucination rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Hallucination Detection - RAG":{"url":"https://huggingface.co/spaces/kluster-ai/LLM-Hallucination-Detection-Leaderboard","metric":"Hallucination rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Public Goods Game":{"url":"https://github.com/lechmazur/goods","metric":"Avg. Contribution (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (AI2D)":{"url":"https://llm-stats.com/benchmarks/ai2d","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":25},"LLM Stats (AIME 2024)":{"url":"https://llm-stats.com/benchmarks/aime-2024","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (AIME 2026)":{"url":"https://llm-stats.com/benchmarks/aime-2026","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (AMC_2022_23)":{"url":"https://llm-stats.com/benchmarks/amc-2022-23","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (APEX-Agents)":{"url":"https://llm-stats.com/benchmarks/apex-agents","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (AndroidWorld_SR)":{"url":"https://llm-stats.com/benchmarks/androidworld-sr","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Artificial Analysis)":{"url":"https://llm-stats.com/benchmarks/artificial-analysis","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (AutomationBench)":{"url":"https://llm-stats.com/benchmarks/automationbench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (BLINK)":{"url":"https://llm-stats.com/benchmarks/blink","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (BabyVision)":{"url":"https://llm-stats.com/benchmarks/babyvision","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Beyond AIME)":{"url":"https://llm-stats.com/benchmarks/beyond-aime","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Bird-SQL (dev))":{"url":"https://llm-stats.com/benchmarks/bird-sql-(dev)","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (BrowseComp Long Context 128k)":{"url":"https://llm-stats.com/benchmarks/browsecomp-long-128k","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (BrowseComp)":{"url":"https://llm-stats.com/benchmarks/browsecomp","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (BrowseComp-zh)":{"url":"https://llm-stats.com/benchmarks/browsecomp-zh","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (C-Eval)":{"url":"https://llm-stats.com/benchmarks/c-eval","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (CC-OCR)":{"url":"https://llm-stats.com/benchmarks/cc-ocr","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (CMMLU)":{"url":"https://llm-stats.com/benchmarks/cmmlu","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (COLLIE)":{"url":"https://llm-stats.com/benchmarks/collie","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (CSimpleQA)":{"url":"https://llm-stats.com/benchmarks/csimpleqa","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (CharXiv-D)":{"url":"https://llm-stats.com/benchmarks/charxiv-d","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (CharXiv-R)":{"url":"https://llm-stats.com/benchmarks/charxiv-r","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (CharadesSTA)":{"url":"https://llm-stats.com/benchmarks/charadessta","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (ChartQA)":{"url":"https://llm-stats.com/benchmarks/chartqa","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Claw-Eval)":{"url":"https://llm-stats.com/benchmarks/claw-eval","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (CodeForces)":{"url":"https://llm-stats.com/benchmarks/codeforces","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (ComplexFuncBench)":{"url":"https://llm-stats.com/benchmarks/complexfuncbench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (CountBench)":{"url":"https://llm-stats.com/benchmarks/countbench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (CyberGym)":{"url":"https://llm-stats.com/benchmarks/cybergym","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (DROP)":{"url":"https://llm-stats.com/benchmarks/drop","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (DeepPlanning)":{"url":"https://llm-stats.com/benchmarks/deep-planning","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (DeepSWE)":{"url":"https://llm-stats.com/benchmarks/deepswe","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (DeepSearchQA)":{"url":"https://llm-stats.com/benchmarks/deepsearchqa","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (DocVQA)":{"url":"https://llm-stats.com/benchmarks/docvqa","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (DocVQAtest)":{"url":"https://llm-stats.com/benchmarks/docvqatest","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (DynaMath)":{"url":"https://llm-stats.com/benchmarks/dynamath","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (ECLeKTic)":{"url":"https://llm-stats.com/benchmarks/eclektic","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (ERQA)":{"url":"https://llm-stats.com/benchmarks/erqa","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (EgoSchema)":{"url":"https://llm-stats.com/benchmarks/egoschema","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (EmbSpatialBench)":{"url":"https://llm-stats.com/benchmarks/embspatialbench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (FLEURS)":{"url":"https://llm-stats.com/benchmarks/fleurs","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Finance Agent)":{"url":"https://llm-stats.com/benchmarks/finance-agent","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (GDP.pdf)":{"url":"https://llm-stats.com/benchmarks/gdp-pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (GDPval-AA)":{"url":"https://llm-stats.com/benchmarks/gdpval-aa","metric":"ELO","theoretical_min":null,"theoretical_max":null},"LLM Stats (Graphwalks BFS <128k)":{"url":"https://llm-stats.com/benchmarks/graphwalks-bfs-<128k","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Graphwalks BFS >128k)":{"url":"https://llm-stats.com/benchmarks/graphwalks-bfs->128k","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Graphwalks parents <128k)":{"url":"https://llm-stats.com/benchmarks/graphwalks-parents-<128k","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Graphwalks parents >128k)":{"url":"https://llm-stats.com/benchmarks/graphwalks-parents->128k","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (HMMT 2025)":{"url":"https://llm-stats.com/benchmarks/hmmt-2025","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (HMMT Feb 26)":{"url":"https://llm-stats.com/benchmarks/hmmt-feb-26","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (HMMT25)":{"url":"https://llm-stats.com/benchmarks/hmmt25","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Hallusion Bench)":{"url":"https://llm-stats.com/benchmarks/hallusion-bench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (HealthBench Hard)":{"url":"https://llm-stats.com/benchmarks/healthbench-hard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (HealthBench Professional)":{"url":"https://llm-stats.com/benchmarks/healthbench-professional","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0,"sources":[{"name":"LLM Stats (HealthBench Professional)","url":"https://llm-stats.com/benchmarks/healthbench-professional","metric":"Score (%)"},{"name":"HealthBench Professional (Anthropic)","url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Length-Adjusted Score (%)"}]},"LLM Stats (HealthBench)":{"url":"https://llm-stats.com/benchmarks/healthbench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (HiddenMath)":{"url":"https://llm-stats.com/benchmarks/hiddenmath","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (IMO-AnswerBench)":{"url":"https://llm-stats.com/benchmarks/imo-answerbench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Include)":{"url":"https://llm-stats.com/benchmarks/include","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (InfoVQA)":{"url":"https://llm-stats.com/benchmarks/infovqa","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (InfoVQAtest)":{"url":"https://llm-stats.com/benchmarks/infovqatest","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Internal API instruction following (hard))":{"url":"https://llm-stats.com/benchmarks/internal-api-instruction-following-(hard)","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (LVBench)":{"url":"https://llm-stats.com/benchmarks/lvbench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Legal Agent Benchmark)":{"url":"https://llm-stats.com/benchmarks/legal-agent-benchmark","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (LongBench v2)":{"url":"https://llm-stats.com/benchmarks/longbench-v2","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (MAXIFE)":{"url":"https://llm-stats.com/benchmarks/maxife","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (MCP Atlas)":{"url":"https://llm-stats.com/benchmarks/mcp-atlas","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (MCP-Mark)":{"url":"https://llm-stats.com/benchmarks/mcp-mark","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (MLVU)":{"url":"https://llm-stats.com/benchmarks/mlvu","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (MLVU-M)":{"url":"https://llm-stats.com/benchmarks/mlvu-m","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (MM-MT-Bench)":{"url":"https://llm-stats.com/benchmarks/mm-mt-bench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (MMBench)":{"url":"https://llm-stats.com/benchmarks/mmbench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (MMBench-V1.1)":{"url":"https://llm-stats.com/benchmarks/mmbench-v1.1","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (MMLU-ProX)":{"url":"https://llm-stats.com/benchmarks/mmlu-prox","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":10},"LLM Stats (MMLU-Redux)":{"url":"https://llm-stats.com/benchmarks/mmlu-redux","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":25},"LLM Stats (MMLongBench-Doc)":{"url":"https://llm-stats.com/benchmarks/mmlongbench-doc","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (MMMLU)":{"url":"https://llm-stats.com/benchmarks/mmmlu","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":25},"LLM Stats (MMStar)":{"url":"https://llm-stats.com/benchmarks/mmstar","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":25},"LLM Stats (MRCR v2 (8-needle))":{"url":"https://llm-stats.com/benchmarks/mrcr-v2-(8-needle)","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (MRCR)":{"url":"https://llm-stats.com/benchmarks/mrcr","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (MVBench)":{"url":"https://llm-stats.com/benchmarks/mvbench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (MathArena Apex)":{"url":"https://llm-stats.com/benchmarks/matharena-apex","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (MathVision)":{"url":"https://llm-stats.com/benchmarks/mathvision","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (MathVista)":{"url":"https://llm-stats.com/benchmarks/mathvista","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (MathVista-Mini)":{"url":"https://llm-stats.com/benchmarks/mathvista-mini","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (MedXpertQA)":{"url":"https://llm-stats.com/benchmarks/medxpertqa","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (MuirBench)":{"url":"https://llm-stats.com/benchmarks/muirbench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Multi-Challenge)":{"url":"https://llm-stats.com/benchmarks/multichallenge","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Multi-IF)":{"url":"https://llm-stats.com/benchmarks/multi-if","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Multi-SWE-Bench)":{"url":"https://llm-stats.com/benchmarks/multi-swe-bench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (MultiPL-E)":{"url":"https://llm-stats.com/benchmarks/multipl-e","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Multilingual MMLU)":{"url":"https://llm-stats.com/benchmarks/multilingual-mmlu","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (NL2Repo)":{"url":"https://llm-stats.com/benchmarks/nl2repo","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (NOVA-63)":{"url":"https://llm-stats.com/benchmarks/nova-63","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Natural Questions)":{"url":"https://llm-stats.com/benchmarks/natural-questions","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Natural2Code)":{"url":"https://llm-stats.com/benchmarks/natural2code","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (OCRBench)":{"url":"https://llm-stats.com/benchmarks/ocrbench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (OCRBench-V2 (en))":{"url":"https://llm-stats.com/benchmarks/ocrbench-v2-(en)","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (OCRBench-V2 (zh))":{"url":"https://llm-stats.com/benchmarks/ocrbench-v2-(zh)","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (OCRBench_V2)":{"url":"https://llm-stats.com/benchmarks/ocrbench-v2","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (ODinW)":{"url":"https://llm-stats.com/benchmarks/odinw","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (OJBench)":{"url":"https://llm-stats.com/benchmarks/ojbench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (OSWorld)":{"url":"https://llm-stats.com/benchmarks/osworld","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (OSWorld-Verified)":{"url":"https://llm-stats.com/benchmarks/osworld-verified","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (OfficeQA Pro)":{"url":"https://llm-stats.com/benchmarks/officeqa-pro","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (OmniDocBench 1.5)":{"url":"https://llm-stats.com/benchmarks/omnidocbench-1.5","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (OpenAI-MRCR: 2 needle 128k)":{"url":"https://llm-stats.com/benchmarks/openai-mrcr:-2-needle-128k","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (OpenAI-MRCR: 2 needle 1M)":{"url":"https://llm-stats.com/benchmarks/openai-mrcr:-2-needle-1m","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (PolyMATH)":{"url":"https://llm-stats.com/benchmarks/polymath","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (PostTrainBench)":{"url":"https://llm-stats.com/benchmarks/posttrainbench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Program Bench)":{"url":"https://llm-stats.com/benchmarks/program-bench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (RealWorldQA)":{"url":"https://llm-stats.com/benchmarks/realworldqa","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (RefCOCO-avg)":{"url":"https://llm-stats.com/benchmarks/refcoco-avg","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (RefSpatialBench)":{"url":"https://llm-stats.com/benchmarks/refspatialbench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (SQuALITY)":{"url":"https://llm-stats.com/benchmarks/squality","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (SWE-Lancer (IC-Diamond subset))":{"url":"https://llm-stats.com/benchmarks/swe-lancer-(ic-diamond-subset)","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (ScreenSpot Pro)":{"url":"https://llm-stats.com/benchmarks/screenspot-pro","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (ScreenSpot)":{"url":"https://llm-stats.com/benchmarks/screenspot","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Seal-0)":{"url":"https://llm-stats.com/benchmarks/seal-0","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (SimpleVQA)":{"url":"https://llm-stats.com/benchmarks/simplevqa","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (SkillsBench)":{"url":"https://llm-stats.com/benchmarks/skillsbench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Social IQa)":{"url":"https://llm-stats.com/benchmarks/social-iqa","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (TAU3-Bench)":{"url":"https://llm-stats.com/benchmarks/tau3-bench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Terminal-Bench 2.1)":{"url":"https://llm-stats.com/benchmarks/terminal-bench-2.1","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (TextVQA)":{"url":"https://llm-stats.com/benchmarks/textvqa","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (TheoremQA)":{"url":"https://llm-stats.com/benchmarks/theoremqa","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Toolathlon)":{"url":"https://llm-stats.com/benchmarks/toolathlon","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (TriviaQA)":{"url":"https://llm-stats.com/benchmarks/triviaqa","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (V*)":{"url":"https://llm-stats.com/benchmarks/v-star","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (VITA-Bench)":{"url":"https://llm-stats.com/benchmarks/vita-bench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Vibe-Eval)":{"url":"https://llm-stats.com/benchmarks/vibe-eval","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Video-MME)":{"url":"https://llm-stats.com/benchmarks/video-mme","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (VideoMME w sub.)":{"url":"https://llm-stats.com/benchmarks/videomme-w-sub.","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (VideoMME w/o sub.)":{"url":"https://llm-stats.com/benchmarks/videomme-w-o-sub.","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (VideoMMMU)":{"url":"https://llm-stats.com/benchmarks/videommmu","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (WMT24++)":{"url":"https://llm-stats.com/benchmarks/wmt24++","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (WideSearch)":{"url":"https://llm-stats.com/benchmarks/widesearch","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (Wild Bench)":{"url":"https://llm-stats.com/benchmarks/wild-bench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (WorldVQA)":{"url":"https://llm-stats.com/benchmarks/worldvqa","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (WritingBench)":{"url":"https://llm-stats.com/benchmarks/writingbench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (ZEROBench)":{"url":"https://llm-stats.com/benchmarks/zerobench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (ZEROBench-Sub)":{"url":"https://llm-stats.com/benchmarks/zerobench-sub","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (ZebraLogic)":{"url":"https://llm-stats.com/benchmarks/zebralogic","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Stats (t2-bench)":{"url":"https://llm-stats.com/benchmarks/t2-bench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Trustworthy - Adversarial":{"url":"https://huggingface.co/spaces/AI-Secure/llm-trustworthy-leaderboard","metric":"Trust Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Trustworthy - Adversarial Demo":{"url":"https://huggingface.co/spaces/AI-Secure/llm-trustworthy-leaderboard","metric":"Trust Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Trustworthy - Ethics":{"url":"https://huggingface.co/spaces/AI-Secure/llm-trustworthy-leaderboard","metric":"Trust Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Trustworthy - Fairness":{"url":"https://huggingface.co/spaces/AI-Secure/llm-trustworthy-leaderboard","metric":"Trust Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Trustworthy - Out-of-Distribution":{"url":"https://huggingface.co/spaces/AI-Secure/llm-trustworthy-leaderboard","metric":"Trust Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Trustworthy - Privacy":{"url":"https://huggingface.co/spaces/AI-Secure/llm-trustworthy-leaderboard","metric":"Trust Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Trustworthy - Stereotype":{"url":"https://huggingface.co/spaces/AI-Secure/llm-trustworthy-leaderboard","metric":"Trust Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Trustworthy - Toxicity":{"url":"https://huggingface.co/spaces/AI-Secure/llm-trustworthy-leaderboard","metric":"Trust Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM Trustworthy Leaderboard":{"url":"https://huggingface.co/spaces/AI-Secure/llm-trustworthy-leaderboard","metric":"Average Trust Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM-AggreFact":{"url":"https://llm-aggrefact.github.io/","metric":"Balanced Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM-WikiRace":{"url":"https://llmwikirace.github.io/","metric":"Average Success (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM-as-a-Reviewer":{"url":"https://benchmarklist.com/benchmarks/llm_as_a_reviewer_benchmarking_their_ability_divergence_and_prompt_injection_resistance_as_paper_reviewers/","metric":"Low (NeurIPS 2022) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LLM2014 Code 2025-07":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/code/2025-07.csv","metric":"Median Score","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Code 2025-09":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/code/2025-09.csv","metric":"Median Score","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Code 2025-09 - C#":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/code/2025-09-lang.csv","metric":"Score","theoretical_min":0.0,"theoretical_max":10.0},"LLM2014 Code 2025-09 - C++":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/code/2025-09-lang.csv","metric":"Score","theoretical_min":0.0,"theoretical_max":10.0},"LLM2014 Code 2025-09 - Golang":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/code/2025-09-lang.csv","metric":"Score","theoretical_min":0.0,"theoretical_max":10.0},"LLM2014 Code 2025-09 - Java":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/code/2025-09-lang.csv","metric":"Score","theoretical_min":0.0,"theoretical_max":10.0},"LLM2014 Code 2025-09 - Python":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/code/2025-09-lang.csv","metric":"Score","theoretical_min":0.0,"theoretical_max":10.0},"LLM2014 Code 2025-09 - TypeScript":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/code/2025-09-lang.csv","metric":"Score","theoretical_min":0.0,"theoretical_max":10.0},"LLM2014 Code 2025-11":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/code/2025-11.csv","metric":"Multi-round Score","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Code 2025-11 - C#":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/code/2025-11-lang.csv","metric":"Score","theoretical_min":0.0,"theoretical_max":10.0},"LLM2014 Code 2025-11 - C++":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/code/2025-11-lang.csv","metric":"Score","theoretical_min":0.0,"theoretical_max":10.0},"LLM2014 Code 2025-11 - Golang":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/code/2025-11-lang.csv","metric":"Score","theoretical_min":0.0,"theoretical_max":10.0},"LLM2014 Code 2025-11 - Java":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/code/2025-11-lang.csv","metric":"Score","theoretical_min":0.0,"theoretical_max":10.0},"LLM2014 Code 2025-11 - Python":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/code/2025-11-lang.csv","metric":"Score","theoretical_min":0.0,"theoretical_max":10.0},"LLM2014 Code 2025-11 - TypeScript":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/code/2025-11-lang.csv","metric":"Score","theoretical_min":0.0,"theoretical_max":10.0},"LLM2014 Logic 2024-04":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2024-04.csv","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2024-05":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2024-05.csv","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2024-06":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2024-06.csv","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2024-07":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2024-07.csv","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2024-08":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2024-08.csv","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2024-09":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2024-09.csv","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2024-10":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2024-10.csv","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2024-12":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2024-12.csv","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2025-01":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2025-01.csv","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2025-03":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2025-03.csv","metric":"Median Score","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2025-04":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2025-04.csv","metric":"Median Score","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2025-05":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2025-05.csv","metric":"Median Score","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2025-06":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2025-06.csv","metric":"Median Score","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2025-08":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2025-08.csv","metric":"Median Score","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2025-09":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2025-09.csv","metric":"Median Score","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2025-10":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2025-10.csv","metric":"Median Score","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2025-11":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2025-11.csv","metric":"Median Score","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2025-12":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2025-12.csv","metric":"Median Score","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2026-01":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2026-01.csv","metric":"Median Score","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2026-02":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2026-02.csv","metric":"Median Score","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2026-03":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2026-03.csv","metric":"Median Score","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2026-04":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2026-04.csv","metric":"Median Score","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2026-05":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2026-05.csv","metric":"Median Score","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2026-06":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2026-06.csv","metric":"Median Score","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Logic 2026-07":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/logic/2026-07.csv","metric":"Median Score","theoretical_min":0.0,"theoretical_max":100.0},"LLM2014 Vision 2025-07":{"url":"https://raw.githubusercontent.com/llm2014/llm_benchmark/main/docs/data/vision/2025-07.csv","metric":"Median Score","theoretical_min":0.0,"theoretical_max":100.0},"LLMEval-Logic Base":{"url":"https://llmeval.com/leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLMEval-Logic Formalization Fixed":{"url":"https://llmeval.com/leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLMEval-Logic Formalization Free":{"url":"https://llmeval.com/leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLMEval-Logic Hard":{"url":"https://llmeval.com/leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLMEval-Logic Hard Sub-Q":{"url":"https://llmeval.com/leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LLMZSZL Leaderboard":{"url":"https://huggingface.co/spaces/amu-cai/LLMZSZL_Leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"LLMsPark - Dictator Game":{"url":"https://llmsparks.github.io/","metric":"Elo Rating","theoretical_min":null,"theoretical_max":null},"LLMsPark - Nim Game":{"url":"https://llmsparks.github.io/","metric":"Elo Rating","theoretical_min":null,"theoretical_max":null},"LLMsPark - Prisoner's Dilemma":{"url":"https://llmsparks.github.io/","metric":"Elo Rating","theoretical_min":null,"theoretical_max":null},"LLMsPark - Trust Game":{"url":"https://llmsparks.github.io/","metric":"Elo Rating","theoretical_min":null,"theoretical_max":null},"LLMsPark - Who Is Spy":{"url":"https://llmsparks.github.io/","metric":"Elo Rating","theoretical_min":null,"theoretical_max":null},"LLaVA-Bench":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score","theoretical_min":null,"theoretical_max":null},"LMArena Preference Proxy":{"url":"https://huggingface.co/spaces/lmarena-ai/preference-proxy-evaluations","metric":"Evaluator Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LMArena Text Arena":{"url":"https://benchmarklist.com/benchmarks/lmsys_chatbot_arena/","metric":"Arena rating (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LMArena WebDev Arena":{"url":"https://benchmarklist.com/benchmarks/lmarena_webdev_arena/","metric":"Arena rating (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LMGame-Bench 2048":{"url":"https://huggingface.co/spaces/lmgame/lmgame_bench","metric":"Score","theoretical_min":0.0,"theoretical_max":null},"LMGame-Bench Ace Attorney":{"url":"https://huggingface.co/spaces/lmgame/lmgame_bench","metric":"Score","theoretical_min":0.0,"theoretical_max":null},"LMGame-Bench Candy Crush":{"url":"https://huggingface.co/spaces/lmgame/lmgame_bench","metric":"Score","theoretical_min":0.0,"theoretical_max":null},"LMGame-Bench Sokoban":{"url":"https://huggingface.co/spaces/lmgame/lmgame_bench","metric":"Score","theoretical_min":0.0,"theoretical_max":null},"LMGame-Bench Super Mario Bros":{"url":"https://huggingface.co/spaces/lmgame/lmgame_bench","metric":"Score","theoretical_min":0.0,"theoretical_max":null},"LMGame-Bench Tetris":{"url":"https://huggingface.co/spaces/lmgame/lmgame_bench","metric":"Score","theoretical_min":0.0,"theoretical_max":null},"LOL Arena":{"url":"https://goodstartlabs.com/leaderboards/lol-arena","metric":"Win Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"LOM Benchmark":{"url":"https://huggingface.co/spaces/Yonyou-AI-Lab/LOM-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LOM Benchmark - BipartiteMatching":{"url":"https://huggingface.co/spaces/Yonyou-AI-Lab/LOM-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LOM Benchmark - CommonNeighbor":{"url":"https://huggingface.co/spaces/Yonyou-AI-Lab/LOM-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LOM Benchmark - CycleDetection":{"url":"https://huggingface.co/spaces/Yonyou-AI-Lab/LOM-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LOM Benchmark - Description":{"url":"https://huggingface.co/spaces/Yonyou-AI-Lab/LOM-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LOM Benchmark - Edge":{"url":"https://huggingface.co/spaces/Yonyou-AI-Lab/LOM-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LOM Benchmark - EntityAttribute":{"url":"https://huggingface.co/spaces/Yonyou-AI-Lab/LOM-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LOM Benchmark - EntityRelation":{"url":"https://huggingface.co/spaces/Yonyou-AI-Lab/LOM-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LOM Benchmark - EntityType":{"url":"https://huggingface.co/spaces/Yonyou-AI-Lab/LOM-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LOM Benchmark - MST":{"url":"https://huggingface.co/spaces/Yonyou-AI-Lab/LOM-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LOM Benchmark - Neighbor":{"url":"https://huggingface.co/spaces/Yonyou-AI-Lab/LOM-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LOM Benchmark - PageRank":{"url":"https://huggingface.co/spaces/Yonyou-AI-Lab/LOM-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LOM Benchmark - Predecessor":{"url":"https://huggingface.co/spaces/Yonyou-AI-Lab/LOM-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LOM Benchmark - RelationBetween":{"url":"https://huggingface.co/spaces/Yonyou-AI-Lab/LOM-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LOM Benchmark - RelationPath":{"url":"https://huggingface.co/spaces/Yonyou-AI-Lab/LOM-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LOM Benchmark - ShortestPath":{"url":"https://huggingface.co/spaces/Yonyou-AI-Lab/LOM-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LOM Benchmark - Summary":{"url":"https://huggingface.co/spaces/Yonyou-AI-Lab/LOM-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LOM Benchmark - TreeCheck":{"url":"https://huggingface.co/spaces/Yonyou-AI-Lab/LOM-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LOM Benchmark - TreeLCA":{"url":"https://huggingface.co/spaces/Yonyou-AI-Lab/LOM-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LVBench":{"url":"https://benchmarklist.com/benchmarks/lvbench/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LaViSA":{"url":"https://benchmarklist.com/benchmarks/lavisa/","metric":"PT-Acc (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LaborBench":{"url":"https://benchmarklist.com/benchmarks/laborbench/","metric":"F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LandmarkBias-3K - City (25 km)":{"url":"https://arxiv.org/html/2607.15255v1","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LandmarkBias-3K - Country (750 km)":{"url":"https://arxiv.org/html/2607.15255v1","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LandmarkBias-3K - Region (200 km)":{"url":"https://arxiv.org/html/2607.15255v1","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Large Language Models Lack Temporal Awareness":{"url":"https://benchmarklist.com/benchmarks/large_language_models_lack_temporal_awareness_of_medical_knowledge/","metric":"Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LatamBoard - ASSIN2 RTE":{"url":"https://huggingface.co/datasets/LatamBoard/leaderboard-results","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LatamBoard - BLUEX":{"url":"https://huggingface.co/datasets/LatamBoard/leaderboard-results","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LatamBoard - ENEM Challenge":{"url":"https://huggingface.co/datasets/LatamBoard/leaderboard-results","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LatamBoard - FLORES Bidirectional":{"url":"https://huggingface.co/datasets/LatamBoard/leaderboard-results","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LatamBoard - FaQuAD NLI":{"url":"https://huggingface.co/datasets/LatamBoard/leaderboard-results","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LatamBoard - OAB Exams":{"url":"https://huggingface.co/datasets/LatamBoard/leaderboard-results","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LatamBoard - Portuguese Score":{"url":"https://huggingface.co/datasets/LatamBoard/leaderboard-results","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LatamBoard - Spanish COPA":{"url":"https://huggingface.co/datasets/LatamBoard/leaderboard-results","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LatamBoard - Spanish Escola":{"url":"https://huggingface.co/datasets/LatamBoard/leaderboard-results","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LatamBoard - Spanish MGSM":{"url":"https://huggingface.co/datasets/LatamBoard/leaderboard-results","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LatamBoard - Spanish OpenBookQA":{"url":"https://huggingface.co/datasets/LatamBoard/leaderboard-results","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LatamBoard - Spanish PAWS":{"url":"https://huggingface.co/datasets/LatamBoard/leaderboard-results","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LatamBoard - Spanish Score":{"url":"https://huggingface.co/datasets/LatamBoard/leaderboard-results","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LatamBoard - Spanish TeleIA":{"url":"https://huggingface.co/datasets/LatamBoard/leaderboard-results","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LatamBoard - Spanish WNLI":{"url":"https://huggingface.co/datasets/LatamBoard/leaderboard-results","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LatamBoard - Spanish XNLI":{"url":"https://huggingface.co/datasets/LatamBoard/leaderboard-results","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LatamBoard - Translation Score":{"url":"https://huggingface.co/datasets/LatamBoard/leaderboard-results","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LatchBio SingleCellBench (Anthropic)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LatchBio SpatialBench (Anthropic)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Lean AI Formalization Leaderboard":{"url":"https://lean-lang.org/eval/","metric":"Solved Problems","theoretical_min":0.0,"theoretical_max":null},"Legal Agent Benchmark (Harvey Held-Out)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"All-Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Legal Agent Benchmark Public Set (Fable/Mythos)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"All-Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Legal Hallucination Detection Benchmark (no ex":{"url":"https://benchmarklist.com/benchmarks/legal_hallucination_detection_benchmark_no_explicit_name_given_but_introduces_a_dataset_of_1_300_brief_excerpts_with_injected_errors/","metric":"F1 (%) - Agentic (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LegalCiteBench":{"url":"https://benchmarklist.com/benchmarks/legalcitebench/","metric":"Overall (Norm.) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LexGenius":{"url":"https://benchmarklist.com/benchmarks/lexgenius/","metric":"CoT average (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LibEvoBench":{"url":"https://benchmarklist.com/benchmarks/libevobench_probing_temporal_knowledge_stratification_in_code_generation_models/","metric":"SEUS (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LinAlg-Bench":{"url":"https://benchmarklist.com/benchmarks/linalg_bench/","metric":"3×3 Overall Model Accuracy (%) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LingOly":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"Exact Match Accuracy","theoretical_min":0.0,"theoretical_max":100.0},"LingOly-TOO":{"url":"https://benchmarklist.com/benchmarks/lingoly_too/","metric":"Obfuscated Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LisanBench":{"url":"https://lisanbench.com/","metric":"Mean Path Length / Current Maximum","theoretical_min":0.0,"theoretical_max":1.0},"LithoBench":{"url":"https://benchmarklist.com/benchmarks/lithobench/","metric":"MCQ$_{\\mathrm{all}}$ (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LiveBench":{"url":"https://benchmarklist.com/benchmarks/livebench/","metric":"LiveBench average (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LiveBench AMPS Hard":{"url":"https://livebench.ai/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"LiveBench Coding Completion":{"url":"https://livebench.ai/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"LiveBench Connections":{"url":"https://livebench.ai/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"LiveBench Cta":{"url":"https://livebench.ai/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"LiveBench LCB Generation":{"url":"https://livebench.ai/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"LiveBench Math Comp":{"url":"https://livebench.ai/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"LiveBench Olympiad":{"url":"https://livebench.ai/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"LiveBench Paraphrase":{"url":"https://livebench.ai/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"LiveBench Plot Unscrambling":{"url":"https://livebench.ai/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"LiveBench Simplify":{"url":"https://livebench.ai/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"LiveBench Story Generation":{"url":"https://livebench.ai/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"LiveBench Summarize":{"url":"https://livebench.ai/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"LiveBench Table Join":{"url":"https://livebench.ai/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"LiveBench Table Reformat":{"url":"https://livebench.ai/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"LiveBench Typos":{"url":"https://livebench.ai/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"LiveBench Web Of Lies V2":{"url":"https://livebench.ai/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"LiveBench Zebra Puzzle":{"url":"https://livebench.ai/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"LiveCodeBench":{"url":"https://livecodebench.github.io/","metric":"Pass@1 avg (%)","theoretical_min":0.0,"theoretical_max":100.0},"LiveCodeBench Pro":{"url":"https://livecodebenchpro.com/","metric":"Rating (CF-style)","theoretical_min":null,"theoretical_max":null},"LiveCodeBench Pro Easy":{"url":"https://livecodebenchpro.com/","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"LiveCodeBench Pro Hard":{"url":"https://livecodebenchpro.com/","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"LiveCodeBench Pro Medium":{"url":"https://livecodebenchpro.com/","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"LiveFMBench":{"url":"https://benchmarklist.com/benchmarks/livefmbench/","metric":"Pass@1 Before (Pre2025) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LiveK12Bench":{"url":"https://benchmarklist.com/benchmarks/livek12bench_have_large_multimodal_models_truly_conquered_high_school_level_examinations/","metric":"Mathematics Acc (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LiveMathematicianBench":{"url":"https://livemathematicianbench.github.io/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LiveMedBench":{"url":"https://zhilingyan.github.io/LiveMedBench/","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LiveOIBench":{"url":"https://liveoibench.github.io/leaderboard.html","metric":"Avg Human Percentile","theoretical_min":0.0,"theoretical_max":100.0},"LiveSQLBench":{"url":"https://benchmarklist.com/benchmarks/livesqlbench/","metric":"Success Rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"LiveSecBench":{"url":"https://livesecbench.intokentech.cn/","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LoHoSearch":{"url":"https://huggingface.co/datasets/meituan-longcat/LoHoSearch","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LogicKor":{"url":"https://lk.instruct.kr/","metric":"Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"LogicKor - Coding":{"url":"https://lk.instruct.kr/","metric":"Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"LogicKor - Grammar":{"url":"https://lk.instruct.kr/","metric":"Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"LogicKor - Math":{"url":"https://lk.instruct.kr/","metric":"Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"LogicKor - Multi-Turn":{"url":"https://lk.instruct.kr/","metric":"Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"LogicKor - Reasoning":{"url":"https://lk.instruct.kr/","metric":"Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"LogicKor - Single-Turn":{"url":"https://lk.instruct.kr/","metric":"Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"LogicKor - Understanding":{"url":"https://lk.instruct.kr/","metric":"Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"LogicKor - Writing":{"url":"https://lk.instruct.kr/","metric":"Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"Logical Consistency Reformulation - Base Accuracy":{"url":"https://arxiv.org/abs/2607.14528","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Long Code Arena":{"url":"https://huggingface.co/spaces/JetBrains-Research/long-code-arena","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Long-Horizon Terminal-Bench":{"url":"https://zli12321.github.io/LHTB/","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LongBench v2":{"url":"https://longbench2.github.io/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LongEmotion":{"url":"https://longemotion.github.io/","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"LongProc":{"url":"https://princeton-pli.github.io/LongProc/","metric":"Accuracy @ 0.5K (%)","theoretical_min":0.0,"theoretical_max":100.0},"LongVideoBench":{"url":"https://longvideobench.github.io/","metric":"Test Total Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"LongWebBench":{"url":"https://benchmarklist.com/benchmarks/longwebbench/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Loong":{"url":"https://github.com/MozerWang/Loong","metric":"Overall Avg Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Lynchmark":{"url":"https://lynchmark.com/","metric":"Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Lyptus Cyber Time Horizons - InterCode-CTF":{"url":"https://lyptusresearch.org/research/gpt-5-5-saturates-offensive-cyber-time-horizons","metric":"pass@1 at 2M tokens (%)","theoretical_min":0.0,"theoretical_max":100.0},"Lyptus Cyber Time Horizons - NL2Bash":{"url":"https://lyptusresearch.org/research/gpt-5-5-saturates-offensive-cyber-time-horizons","metric":"pass@1 at 2M tokens (%)","theoretical_min":0.0,"theoretical_max":100.0},"MASLegalBench":{"url":"https://benchmarklist.com/benchmarks/maslegalbench/","metric":"Best accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MATH Level 5":{"url":"https://epoch.ai/benchmarks/math-level-5","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MATH-MC Level 1":{"url":"https://github.com/Geralt-Targaryen/MC-Evaluation","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MATH-MC Level 2":{"url":"https://github.com/Geralt-Targaryen/MC-Evaluation","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MATH-MC Level 3":{"url":"https://github.com/Geralt-Targaryen/MC-Evaluation","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MATH-MC Level 4":{"url":"https://github.com/Geralt-Targaryen/MC-Evaluation","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MATH-MC Level 5":{"url":"https://github.com/Geralt-Targaryen/MC-Evaluation","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MATH-Perturb (Hard)":{"url":"https://math-perturb.github.io/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MAVEN-Bench":{"url":"https://benchmarklist.com/benchmarks/maven_bench/","metric":"Acc. (%) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MBPP+":{"url":"https://benchmarklist.com/benchmarks/mbpp/","metric":"MBPP+ pass@1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MCJudgeBench":{"url":"https://benchmarklist.com/benchmarks/mcjudgebench_a_benchmark_for_constraint_level_judge_evaluation_in_multi_constraint_instruction_following/","metric":"CJAR (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MCP Atlas":{"url":"https://benchmarklist.com/benchmarks/mcp_atlas/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MCP-Atlas (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MCP-Universe":{"url":"https://benchmarklist.com/benchmarks/mcp_universe/","metric":"Overall Success Rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MCP-Universe (LLM w/ Function Calls)":{"url":"https://mcp-universe.github.io/","metric":"Avg Evaluator Score","theoretical_min":0.0,"theoretical_max":100.0},"MCP-Universe (LLM w/ ReAct)":{"url":"https://mcp-universe.github.io/","metric":"Avg Evaluator Score","theoretical_min":0.0,"theoretical_max":100.0},"MCPBench":{"url":"https://huggingface.co/spaces/mcpbench/mcp-bench","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MCPMark":{"url":"https://mcpmark.ai/leaderboard","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":null},"MDPBench":{"url":"https://huggingface.co/datasets/Delores-Lin/MDPBench","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Application: Coding":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Application: Information Extraction":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Application: Knowledge":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Application: Mathematics":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Application: Metrics":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Application: Perception":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Application: Planning":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Application: Science":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Core":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Macro Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Input Count: 1 Image":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Input Count: 2-3 Images":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Input Count: 4-5 Images":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Input Count: 6-8 Images":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Input Count: 9 Images Or More":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Input Count: Video":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Input Format: 3D Models And Aerial Imagery":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Input Format: Artistic And Creative Content":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Input Format: Diagrams And Data Visualizations":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Input Format: Photographs":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Input Format: Text Based Images And Documents":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Input Format: User Interface Screenshots":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Input Format: Videos":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Open":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Macro Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Output Format: Contextual Formatted Text":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Output Format: Exact Text":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Output Format: Multiple Choice":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Output Format: Numerical Data":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Output Format: Open Ended Output":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Output Format: Structured Output":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Skill: Commonsense And Social Reasoning":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Skill: Domain Specific Knowledge And Skills":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Skill: Ethical And Safety Reasoning":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Skill: Language Understanding And Generation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Skill: Mathematical And Logical Reasoning":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Skill: Object Recognition And Classification":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Skill: Planning And Decision Making":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Skill: Scene And Event Understanding":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Skill: Spatial And Temporal Reasoning":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench - Skill: Text Recognition (OCR)":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - 2D Image Jigsaw Puzzle Easy":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - 3D Fragments Understanding":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - 3D Indoor Scene Text BBox Prediction":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - 3D Indoor Scene Text BBox Selection":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - ARC AGI":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - ASCII Art 30":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - ASCII Art Understanding":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - AV Human Multiview Counting":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - AV Multicamera Tracking Predict BBox":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - AV Vehicle Multiview Counting":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - AV View Identification":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Action Prediction":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Action Sequence":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Action Sequence Understanding":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - ActivityNetQA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Actor Recognition In Movie":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Ad Count Detection":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Adapted CVBench Count":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Adapted CVBench Depth":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Adapted CVBench Distance":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Adapted CVBench Relation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Algebra":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Ancient Map Understanding":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Animal Pose Estimation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Annoying Word Search":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Interactive Operations Alipay":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Interactive Operations Amazon":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Interactive Operations Excel":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Interactive Operations Instagram":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Interactive Operations LeetCode":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Interactive Operations Notes":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Interactive Operations PPT":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Interactive Operations TikTok":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Interactive Operations Twitter":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Interactive Operations Word":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Interactive Operations YouTube":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Interactive Operations Zoom":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Interactive Operations iPhone Settings":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Layout Understanding Alipay":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Layout Understanding Amazon":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Layout Understanding Excel":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Layout Understanding Instagram":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Layout Understanding LeetCode":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Layout Understanding Notes":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Layout Understanding PPT":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Layout Understanding TikTok":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Layout Understanding Twitter":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Layout Understanding Word":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Layout Understanding YouTube":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Layout Understanding Zoom":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - App Layout Understanding iPhone Settings":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Art Explanation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Autonomous Driving Scene Analysis":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Autorater 3D Model Texturing":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Autorater Aesthetics":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Autorater Artifact":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Autorater Artifact Reason":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Autorater Control":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Autorater Mask":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Autorater Motion Guided Editing":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Autorater Semantics":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Autorater Subject":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Autorater Unmask":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Average Humidity Estimate Plot":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Background Change":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Ball Cup Swap 3":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Bar Chart Interpretation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Biology Exams V":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Bongard Problem":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Booking Web Rating":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Booking Web Recommendation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Brand Logo Recognition And Elaboration":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Bridge Strategies Advanced":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Bridge Strategies Expert":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Bridge Strategies World-Class":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - CLEVR Arithmetic":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - CLEVRER Moving Direction Video":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - CLEVRER Object Existence Video":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - CLEVRER Physics":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - CLEVRER Video Moving Object Count":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - CLEVRER Video Moving Object Property Recognition":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - COCO OOD Global Image Retrieval By Query Property":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - COCO Object Detection By Query Property":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - COCO Person Detection":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - CVBench Adapted CVBench Count":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - CVBench Adapted CVBench Depth":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - CVBench Adapted CVBench Distance":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - CVBench Adapted CVBench Relation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Calendar Schedule Suggestion":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Camera Trajectory To Video Selection":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Character Recognition In TV Shows":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Chart VQA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Cheapest Flight Identification":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Chemistry Exams V":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Chess Find Legal Moves":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Chess Puzzle Single Step":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Chess Puzzles Checkmate":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Chess Puzzles Crushing":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Chess Puzzles Equality":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Chess Syzygy Endgames":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Chess Winner Identification":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Chinese Idiom Recognition":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Circuit Diagram Understanding":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Clip Stable Diffusion Generate":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Code Add Tag":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Code Error Line Identification":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Code Execution":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Code Match Problem":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Code Output Result":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Code Programming Extremely Hard":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Code Programming Test Advanced":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Code Programming Test Easy":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Code Programming Test Hard":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Code Retrieval":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Code Solution Compare":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Code Translation Advanced":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Code Translation Easy":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Code Translation Hard":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Code Translation Python":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Code Visualization Output Understanding":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Comic Page Ordering":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Constrained Generation Contain Contain Images":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Constrained Generation Contain Contain Length":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Constrained Generation Contain Length":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Constrained Generation Contain Position Images":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Constrained Generation Contain Position Length":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Constrained Generation Contain Repeat Length":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Constrained Generation Multi Contain Position Only":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Constrained Generation Multi Contain Repeat":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Constrained Generation Multi Contain Repeat Position Only Length":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Constrained Generation Xor Images":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Contain Contain Images":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Contain Contain Length":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Contain Length":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Contain Position Images":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Contain Position Length":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Contain Repeat Length":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Counterfactual Arithmetic":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Counting":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Counting Multi Image":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Counting Single Image":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Crossword Mini 5x5":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Cultural VQA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - DOCCI Image Description Long":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - DVQA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Deciphering Oracle Bone":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Defeasible Reasoning":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Dish Ingredient Match":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Distinguish AI Generated Image":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Doc VQA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Egocentric Analysis Single Image":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Egocentric Spatial Reasoning":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Electricity Future Prediction From Table":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Electricity Load Estimate Plot":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Electricity Plot Future Prediction":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Electrocardiogram":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Emotion Recognition":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Entertainment Web Game Style":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Exchange Rate Estimate Plot":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Extract Webpage Headline":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - FUNSD Document QA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Face Attribute Edit":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Face Identity Matching":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Face Keypoint Detection":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Face Swap":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Famous Building Recognition":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Figurative Speech Explanation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - FigureQA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Finance Table Understanding":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Flowchart Code Generation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Font Recognition":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Forensic Detection Of Different Images":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - FunQA Unexpected Action Creative Video":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - FunQA Unexpected Action Humor Video":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - FunQA Unexpected Action Magic Video":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Functionality Matching In Different Objects":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Funny Image Title":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - GUI Act Mobile Swipe":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - GUI Act Mobile Tap":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - GUI Act Web Multi":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - GUI Act Web Single":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - GUI Chat Easy":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - GUI Chat Hard":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Game Info Parsing":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Game Info Retrieval":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Game Platform Support Identification":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Generated Video Artifacts":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Geographic Remote Sensing Land Cover":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Geometry Analytic":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Geometry Area":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Geometry Descriptive":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Geometry Length":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Geometry Plot Position Relationship":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Geometry Reasoning Circled Letter":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Geometry Reasoning Count Line Intersections":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Geometry Reasoning Grid":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Geometry Reasoning Nested Squares":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Geometry Reasoning Overlapped Circle":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Geometry Solid":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Geometry Transformation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Go Capture Stone":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Google Street View Circle Reasoning":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Google Street View Circle Sorting":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Google Street View Direction Understanding":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Google Street View Line Reasoning":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Google Street View Line Sorting":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Graph Chordless Cycle":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Graph Connectivity":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Graph Hamiltonian Cycle":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Graph Hamiltonian Path":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Graph Interpretation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Graph Isomorphism":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Graph Max Flow":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Graph Shortest Path Kamada-Kawai":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Graph Shortest Path Planar":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Graph Theory":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Guess Image Generation Prompt":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Handwritten Math Expression Extraction":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Hashtag Recommendation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Healthcare Info Judgement":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Highest Discount Game Price Identification":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Hotel Booking Confirmation Parsing":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Human Relationship Reasoning":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Humor Explanation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Humor Understand Caption Match":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - IAM Line OCR And Locate":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - ILI Ratio Future Prediction":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - IQ Test":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - IQ Test Open Ended":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Icon Arithmetic Puzzle":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - IconQA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - IconQA Count And Reasoning":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Image Captioning With Additional Requirements":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Image Humor Understanding":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Image Style Recognition":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Image Translation EN-to-CN":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Insect Order Classification":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Interpret Force Perspective Illusion":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Ishihara Test":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - KVQA Knowledge Aware QA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Knowledge Graph Understanding":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Knowledge Sign Recognition":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - LLaVAGuard":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - LaTeX Complex Formula Conversion":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Landmark Check Two Images":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Landmark Recognition And QA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - License Plate Recognition":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Location VQA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Logical Reasoning 2D Folding":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Logical Reasoning 2D Views Of 3D Shapes":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Logical Reasoning Find Odd One Out":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Logical Reasoning Fit Pattern":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Logo2K Same Type Logo Retrieval":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Long String Letter Recognition":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Long String Number Recognition":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - MFC Bench Check Background Change":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - MFC Bench Check Clip Stable Diffusion Generate":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - MFC Bench Check Face Attribute Edit":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - MFC Bench Check Face Swap":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - MFC Bench Check Out Of Context":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - MFC Bench Check Text Entity Replace":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - MFC Bench Check Text Style":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - MFC Bench Check Veracity":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - MMMU Physics Chemistry MCQ":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - MMMU Physics Chemistry Selected":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - MMMU Pro Exam Screenshot":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - MMSoc HatefulMemes":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - MMSoc Memotion":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - MMSoc Misinformation GossipCop":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - MMSoc Misinformation PolitiFact":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - MNIST Pattern":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - MVSA Sentiment Classification":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Mahjong":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Map Diagram QA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Math Breakpoint":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Math Convexity Value Estimation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Math Exams V":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Math Parity":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Maze 2D 8x8":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Media Homepage Profile":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Media QA Web Stack Overflow":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Media Recommend Solutions Stack Overflow":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Medical Abdomen Endoscopy Organ Recognition":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Medical Abdomen MRI Organ Recognition":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Medical Blood Vessels Recognition":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Medical Cell Recognition":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Medical Content Based Retrieval Radiology":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Medical Counting Lymphocytes":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Medical Image Artifacts Identification":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Medical Keywords Based Retrieval Non Radiology":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Medical Multi Organ Segmentation Rater":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Medical Parasite Detection":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Medical Polyp Segmentation Single Object Rater":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Medical Retrieval Given Surgeon Activity":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Meme Explain":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Memorization Chinese Celebrity":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Memorization Famous Treaty":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Memorization Indian Celebrity":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Memorization Papers":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Mensa IQ Test":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Mindmap Elements Parsing":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Monthly Weather Days Count":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Move Position To Position Hanoi 4 Pole":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Movie Info Parsing":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Movie Info Retrieval":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Movie Retrieval By Actor":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - MuMA Theory Of Mind Belief Of Goal":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - MuMA Theory Of Mind Social Goal":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multi Contain Position Only":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multi Contain Repeat":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multi Contain Repeat Position Only Length":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multi Load Type Prediction From Plot":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multilingual Game Info Parsing":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multilingual Manual Explanation Scooter Arabic":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multilingual Manual Explanation Scooter Chinese":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multilingual Manual Explanation Scooter French":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multilingual Manual Explanation Scooter Russian":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multilingual Manual Explanation Scooter Spanish":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multilingual Movie Info Parsing":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multilingual News QA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multilingual Ruozhiba Explanation Arabic":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multilingual Ruozhiba Explanation English":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multilingual Ruozhiba Explanation French":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multilingual Ruozhiba Explanation Japanese":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multilingual Ruozhiba Explanation Russian":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multilingual Ruozhiba Explanation Spanish":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multiple States Identify Africa":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multiple States Identify Americas":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multiple States Identify Asia":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multiple States Identify Europe":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Multiview Reasoning Camera Moving":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Music Info Parsing":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Music Info Retrieval":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Music Sheet Author":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Music Sheet Format QA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Music Sheet Name":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Music Sheet Note Count":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Music Sheet Sentiment":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - NExT-QA MC":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - NExT-QA OE":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - NLVR2 Two Image Compare QA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Newspaper OCR In Query Box":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Newspaper Page Parse And Count":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Next Action Prediction":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Number Comparison":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Number Puzzle Kakuro 5x5":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Number Puzzle Sudoku":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - OCR Article Authors":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - OCR Article Journal":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - OCR Math Equation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - OCR Math Math":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - OCR Math Text LaTeX":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - OCR Math TheoremQA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - OCR Open Ended QA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - OCR Resume Employer Plain":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - OCR Resume Experience Plain":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - OCR Resume School Plain":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - OCR Resume Skill Plain":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - OCR Table To CSV":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - OCR Table To HTML":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - OCR Table To LaTeX":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - OCR Table To Markdown":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - OCRQA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Orchestra Score Recognition":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Out Of Context":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - PMC VQA Medical Image QA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Painting QA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Panel Images Multi Question":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Panel Images Single Question":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Paper Review Acceptance":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Paper Review Rating":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Paper Review Writing":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Paper VQA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Perception Test Object Shuffle Video":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Perception Test Video Action Count":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Perception Test Video Character Order":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Photo Sharing Image Retrieval":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Photoshop Operation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Physical Property Reasoning":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Physics Exams V":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Pictionary Cartoon Drawing Guess":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Pictionary Chinese Food Image-to-EN":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Pictionary Doodle Guess":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Pictionary GenAI Output Chinese":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Pictionary Skribbl.io":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Places365 Scene Type Classification":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Places365 Similar Scene Retrieval":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Planning Screenshot Barman":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Planning Screenshot Blocksworld":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Planning Screenshot Floortile":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Planning Screenshot Grippers":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Planning Screenshot Storage":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Planning Screenshot Termes":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Planning Screenshot Tyreworld":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Planning Visual Barman":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Planning Visual Blocksworld":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Planning Visual Floortile":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Planning Visual Grippers":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Planning Visual Storage":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Planning Visual Termes":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Play Go Capture Stone":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Poetry Acrostic":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Poetry Acrostic Alliteration":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Poetry Custom Rhyming Scheme":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Poetry Haiku":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Poetry Limerick":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Poetry Petrarchian Sonnet Optional Meter":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Poetry Shakespearean Sonnet":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Pokemon 3D Recognition":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Polygon Interior Angles":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Position Relationship":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Product OCR QA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Question Solution Solving":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Quizlet Question Solving":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - RealWorld QA EN-to-CN":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Rebus":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Recipe Image Ordering":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Recover Masked Word In Figure":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Red Teaming Captcha":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Red Teaming Celebrity":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Red Teaming Jailbreak":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Red Teaming Politics":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Red Teaming Racial":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Red Teaming Visual Misleading":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Red Teaming Visual Order A":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Red Teaming Visual Order B":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Relative Depth Of Different Points":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Relative Reflectance Of Different Regions":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Remaining Playback Time Calculation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Research Website Parsing Blog Post":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Research Website Parsing Homepage":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Research Website Parsing Publication":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Reward Models I2T Reward":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Reward Models T2I Reward":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Road Map Find Highway Between Two Place":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Rocks Samples Compare":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Rocks Samples Identify":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - SNLI-VE Visual Entailment":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - STA Action Localization Video":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - SceneQA Scene Transition Video":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - SciBench Calculus Without Solution":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - SciBench Fundamental Without Solution":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - SciBench With Solution Open Ended":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Science Basic Physics":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Science Figure Explanation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Science Molecule Chemistry":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Screenshot LightEval Math":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Screenshot TheoremQA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Semantic Matching Of Two Images":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Shape Composition Colours":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Shape Composition Shapes":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Sign Language":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Signage Navigation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Signboard Identification":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Single Person Pose Estimation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Soccer Offside":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Song Title Identification From Lyrics":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Stack Overflow Debug QA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Star Object Interaction Video":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Stock Info Parsing":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Stock Info Retrieval":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Stock Price Future Prediction":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Super CLEVR":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Super CLEVR Scene Understanding":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Symbolic Graphics Programs Computer Aided Design":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Symbolic Graphics Programs Scalable Vector Graphics":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - TI Fused VQA Biology":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - TI Fused VQA Chemistry":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - TI Fused VQA Math":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - TI Fused VQA Physics":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - TQA Textbook QA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - TV Show Info Parsing":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - TV Show Retrieval By Character":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Table Understanding":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Table Understanding Complex Question Answering":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Table Understanding Fact Verification":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Table Understanding FeTaQA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Table-to-LaTeX Complex":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Text Entity Replace":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Text Style":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Top Rated Hotel Identification":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Top Video Creator Identification":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Topological Sort":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Traffic Accident Analysis":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Traffic Future Prediction From Line Plot":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Trance Physics Reasoning Basic":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Trance Physics Reasoning Event":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Trance Physics Reasoning View":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Transit Map Intersection Points":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Tweets Captioning":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Unusual Images":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - VLN English Next Step":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - VLN Hindi Next Step":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - VLN Identify Location":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - VLN Identify Robot":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - VLN Telugu Next Step":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - VLNQA Egocentric Navigation Video":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Veracity":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Vibe Eval":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Vibe Eval Open":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Vibe Eval Phrase":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Vibe Eval Short Phrase":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Video Action Recognition":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Video Camera Motion Description":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Video Content Follow Up":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Video Content Reasoning":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Video Detail Description":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Video Eval Dynamic Pref":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Video Eval Factual Pref":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Video Eval Visual Pref":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Video Grounding Spatial":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Video Grounding Temporal":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Video Intent Recognition":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Video Motion Matching 3D Real":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Video Motion Matching Real 3D":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Video QA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Video Segments Reordering":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Video Short Title":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Video Summary":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Video To Camera Trajectory Retrieval":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Video2Notes":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - VisDial Visual Dialog Image Guessing":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Visual Correspondence In Two Images":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Visual Dialog Image Guessing":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Visual Prediction Rater 3D Assembled Quality Understanding":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Visual Prediction Rater Depth Estimation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Visual Prediction Rater Novel View Synthesis":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Visual Prediction Rater Openable Part Segmentation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Visual Prediction Rater Panoptic Segmentation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Visual Prediction Rater Plane Segmentation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Visual Prediction Rater Semantic Segmentation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Visual Prediction Rater Surface Normal Estimation":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Visualization With Code":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - VizWiz Quality Assessment For Blind":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - WIDER FACE Face Count And Event Classification":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Waldo":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Waybill Number Sequence Extraction":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Weather Info Parsing":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Weather Info Retrieval":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Weather Map Climate Type Temperature Parsing":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Web Action Grounding":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Web Action Prediction":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Webpage Code Understanding":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - WikiHow Complex Task Completion":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Worldle":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - Xor Images":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - YouTube Video Info Parsing":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MEGA-Bench Task - arXiv VQA":{"url":"https://huggingface.co/spaces/TIGER-Lab/MEGA-Bench","metric":"Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"METR Benchmark":{"url":"https://metr.org/time-horizons/","metric":"50% Time Horizon (hours)","theoretical_min":0.0,"theoretical_max":null},"METR Benchmark (80% Horizon)":{"url":"https://metr.org/time-horizons/","metric":"80% Time Horizon (hours)","theoretical_min":0.0,"theoretical_max":null},"MHGraphBench":{"url":"https://benchmarklist.com/benchmarks/mhgraphbench_knowledge_graph_grounded_benchmarking_of_mental_health_knowledge_in_large_language_models/","metric":"Avg$_{All^\\ast}$ (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MIA-Bench":{"url":"https://huggingface.co/spaces/opencompass/openlmm_subjective_leaderboard","metric":"Overall Score","theoretical_min":0.0,"theoretical_max":100.0},"MILU":{"url":"https://benchmarklist.com/benchmarks/milu/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MIRAGE":{"url":"https://benchmarklist.com/benchmarks/mirage/","metric":"Violent completion rate, $C_1$ (%) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MLB":{"url":"https://benchmarklist.com/benchmarks/mlb_clinical_applications/","metric":"Average (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MLS-Bench Lite":{"url":"https://mls-bench.com/leaderboard","metric":"Score","theoretical_min":null,"theoretical_max":null},"MLX Benchmark V2":{"url":"https://huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MLX Benchmark V2 - Coding":{"url":"https://huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MLX Benchmark V2 - Conceptual":{"url":"https://huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MLX Benchmark V2 - Debugging":{"url":"https://huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MLX Benchmark V2 - Easy":{"url":"https://huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MLX Benchmark V2 - Fill Blank":{"url":"https://huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MLX Benchmark V2 - Hard":{"url":"https://huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MLX Benchmark V2 - MLX Core":{"url":"https://huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MLX Benchmark V2 - MLX Embeddings":{"url":"https://huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MLX Benchmark V2 - MLX Embeddings LoRA":{"url":"https://huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MLX Benchmark V2 - MLX LM":{"url":"https://huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MLX Benchmark V2 - MLX LM LoRA":{"url":"https://huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MLX Benchmark V2 - MLX NN":{"url":"https://huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MLX Benchmark V2 - MLX Optimizers":{"url":"https://huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MLX Benchmark V2 - MLX VLM":{"url":"https://huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MLX Benchmark V2 - Medium":{"url":"https://huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MLX Benchmark V2 - Multiple Choice":{"url":"https://huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MLX Benchmark V2 - QA":{"url":"https://huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MLX Benchmark V2 - True/False":{"url":"https://huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MLX Benchmark V2 - Very Hard":{"url":"https://huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MM-IFEval":{"url":"https://huggingface.co/spaces/opencompass/openlmm_subjective_leaderboard","metric":"Overall Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MM-IssueLoc - File Localization":{"url":"https://arxiv.org/html/2607.15205v1","metric":"File Acc@5 (%)","theoretical_min":0.0,"theoretical_max":100.0},"MM-IssueLoc - Function Localization":{"url":"https://arxiv.org/html/2607.15205v1","metric":"Function Acc@10 (%)","theoretical_min":0.0,"theoretical_max":100.0},"MM-Vet":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0,"sources":[{"name":"MM-Vet","url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score"},{"name":"OpenCompass Multimodal - MM-Vet","url":"https://rank.opencompass.org.cn","metric":"Score (%)"}]},"MMAU":{"url":"https://benchmarklist.com/benchmarks/mmau/","metric":"Avg Test (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MMAlignBench":{"url":"https://huggingface.co/spaces/opencompass/openlmm_subjective_leaderboard","metric":"Overall Reward","theoretical_min":-100.0,"theoretical_max":100.0},"MMBench V1.1":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMCL-Bench":{"url":"https://benchmarklist.com/benchmarks/mmcl_bench/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MME-CC":{"url":"https://randomtutu.github.io/MME-CC/","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MME-Reasoning":{"url":"https://alpha-innovator.github.io/mmereasoning.github.io/","metric":"Overall Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMGist":{"url":"https://benchmarklist.com/benchmarks/mmgist/","metric":"Macro ↑ (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MMLU":{"url":"https://epoch.ai/benchmarks/mmlu","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":25},"MMLU-by-task - ARC Challenge":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Abstract Algebra":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Anatomy":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Astronomy":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Business Ethics":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Clinical Knowledge":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - College Biology":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - College Chemistry":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - College Computer Science":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - College Mathematics":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - College Medicine":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - College Physics":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Computer Security":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Conceptual Physics":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Econometrics":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Electrical Engineering":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Elementary Mathematics":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Formal Logic":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Global Facts":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - HellaSwag":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - High School Biology":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - High School Chemistry":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - High School Computer Science":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - High School European History":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - High School Geography":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - High School Government And Politics":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - High School Macroeconomics":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - High School Mathematics":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - High School Microeconomics":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - High School Physics":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - High School Psychology":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - High School Statistics":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - High School US History":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - High School World History":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Human Aging":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Human Sexuality":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - International Law":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Jurisprudence":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Logical Fallacies":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Machine Learning":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Management":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Marketing":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Medical Genetics":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Miscellaneous":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Moral Disputes":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Moral Scenarios":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Nutrition":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Philosophy":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Prehistory":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Professional Accounting":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Professional Law":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Professional Medicine":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Professional Psychology":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Public Relations":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Security Studies":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Sociology":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - TruthfulQA MC1":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - TruthfulQA MC2":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - US Foreign Policy":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - Virology":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task - World Religions":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLU-by-task Leaderboard":{"url":"https://huggingface.co/spaces/CoreyMorris/MMLU-by-task-Leaderboard","metric":"MMLU Average (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMLongBench-Doc - Accuracy":{"url":"https://huggingface.co/spaces/OpenIXCLab/mmlongbench-doc","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMMLU (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMMU Benchmark":{"url":"https://mmmu-benchmark.github.io/","metric":"Validation Score","theoretical_min":0.0,"theoretical_max":100.0},"MMMU-Pro No Tools (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMMU-Pro With Tools (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMNeedle (1 Image, 2*2 Stitching, Exact Accuracy)":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"1 Image, 2*2 Stitching, Exact Accuracy","theoretical_min":0.0,"theoretical_max":100.0},"MMNeedle (1 Image, 4*4 Stitching, Exact Accuracy)":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"1 Image, 4*4 Stitching, Exact Accuracy","theoretical_min":0.0,"theoretical_max":100.0},"MMNeedle (1 Image, 8*8 Stitching, Exact Accuracy)":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"1 Image, 8*8 Stitching, Exact Accuracy","theoretical_min":0.0,"theoretical_max":100.0},"MMNeedle (10 Images, 1*1 Stitching, Exact Accuracy)":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"10 Images, 1*1 Stitching, Exact Accuracy","theoretical_min":0.0,"theoretical_max":100.0},"MMNeedle (10 Images, 2*2 Stitching, Exact Accuracy)":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"10 Images, 2*2 Stitching, Exact Accuracy","theoretical_min":0.0,"theoretical_max":100.0},"MMNeedle (10 Images, 4*4 Stitching, Exact Accuracy)":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"10 Images, 4*4 Stitching, Exact Accuracy","theoretical_min":0.0,"theoretical_max":100.0},"MMNeedle (10 Images, 8*8 Stitching, Exact Accuracy)":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"10 Images, 8*8 Stitching, Exact Accuracy","theoretical_min":0.0,"theoretical_max":100.0},"MMSI-Bench":{"url":"https://github.com/InternRobotics/MMSI-Bench","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMStar":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"sources":[{"name":"MMStar","url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)"},{"name":"OpenCompass Multimodal - MMStar","url":"https://rank.opencompass.org.cn","metric":"Score (%)"}],"random_baseline":25},"MMTU":{"url":"https://huggingface.co/spaces/MMTU-benchmark/mmtu-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMTU - Column Relationship":{"url":"https://huggingface.co/spaces/MMTU-benchmark/mmtu-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMTU - Column Transform":{"url":"https://huggingface.co/spaces/MMTU-benchmark/mmtu-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMTU - Data Cleaning":{"url":"https://huggingface.co/spaces/MMTU-benchmark/mmtu-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMTU - KB mapping":{"url":"https://huggingface.co/spaces/MMTU-benchmark/mmtu-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMTU - NL-2-code":{"url":"https://huggingface.co/spaces/MMTU-benchmark/mmtu-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMTU - Table Join":{"url":"https://huggingface.co/spaces/MMTU-benchmark/mmtu-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMTU - Table Matching":{"url":"https://huggingface.co/spaces/MMTU-benchmark/mmtu-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMTU - Table QA":{"url":"https://huggingface.co/spaces/MMTU-benchmark/mmtu-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMTU - Table Transform":{"url":"https://huggingface.co/spaces/MMTU-benchmark/mmtu-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMTU - Table Understanding":{"url":"https://huggingface.co/spaces/MMTU-benchmark/mmtu-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMToM-QA":{"url":"https://chuanyangjin.com/mmtom-qa-leaderboard","metric":"All (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMToM-QA - Belief":{"url":"https://chuanyangjin.com/mmtom-qa-leaderboard","metric":"Belief (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMToM-QA - Goal":{"url":"https://chuanyangjin.com/mmtom-qa-leaderboard","metric":"Goal (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMVU":{"url":"https://benchmarklist.com/benchmarks/mmvu/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MMedBench":{"url":"https://henrychur.github.io/MultilingualMedQA/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MMedBench - Rationale (BLEU-1)":{"url":"https://henrychur.github.io/MultilingualMedQA/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MRCR-v2 128k":{"url":"https://benchmarklist.com/benchmarks/mrcr_v2_128k/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MRR-Benchmark":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"Total Column Score","theoretical_min":0.0,"theoretical_max":null},"MT-Bench PL - Coding":{"url":"https://huggingface.co/spaces/speakleash/mt-bench-pl","metric":"Judge Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"MT-Bench PL - Extraction":{"url":"https://huggingface.co/spaces/speakleash/mt-bench-pl","metric":"Judge Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"MT-Bench PL - Humanities":{"url":"https://huggingface.co/spaces/speakleash/mt-bench-pl","metric":"Judge Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"MT-Bench PL - Math":{"url":"https://huggingface.co/spaces/speakleash/mt-bench-pl","metric":"Judge Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"MT-Bench PL - Overall":{"url":"https://huggingface.co/spaces/speakleash/mt-bench-pl","metric":"Judge Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"MT-Bench PL - Reasoning":{"url":"https://huggingface.co/spaces/speakleash/mt-bench-pl","metric":"Judge Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"MT-Bench PL - Roleplay":{"url":"https://huggingface.co/spaces/speakleash/mt-bench-pl","metric":"Judge Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"MT-Bench PL - STEM":{"url":"https://huggingface.co/spaces/speakleash/mt-bench-pl","metric":"Judge Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"MT-Bench PL - Writing":{"url":"https://huggingface.co/spaces/speakleash/mt-bench-pl","metric":"Judge Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"MT-JailBench":{"url":"https://benchmarklist.com/benchmarks/mt_jailbench_a_modular_benchmark_for_understanding_multi_turn_jailbreak_attacks/","metric":"CrescendoX ASR (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MUSE":{"url":"https://benchmarklist.com/benchmarks/muse_benchmarking_manufacturable_functional_and_assemblable_text_to_cad_generation/","metric":"Final Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MUTATE":{"url":"https://benchmarklist.com/benchmarks/mutate/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MWS Vision Bench":{"url":"https://huggingface.co/datasets/MTSAIR/MWS-Vision-Bench","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MageBench":{"url":"https://mage-bench.com/season/0/rankings/","metric":"Rating","theoretical_min":null,"theoretical_max":null},"MageBench S1":{"url":"https://mage-bench.com/season/1/rankings/","metric":"Rating","theoretical_min":null,"theoretical_max":null},"MageBench S2":{"url":"https://mage-bench.com/season/2/rankings/","metric":"Rating","theoretical_min":null,"theoretical_max":null},"MageBench Season 1":{"url":"https://benchmarklist.com/benchmarks/mage_bench_season1/","metric":"Rating (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Magis-Bench":{"url":"https://benchmarklist.com/benchmarks/magis_bench/","metric":"AVG (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Managing Procedural Memory in LLM Agents":{"url":"https://benchmarklist.com/benchmarks/managing_procedural_memory_in_llm_agents_control_adaptation_and_evaluation/","metric":"size (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Math-VR":{"url":"https://benchmarklist.com/benchmarks/math_vr/","metric":"Overall Answer Correctness (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MathArena - AIME 2025":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=aime--aime_2025&amp%3Bcomp=aime--aime_2025&comp=aime--aime_2025","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - AIME 2026":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=aime--aime_2026&amp%3Bcomp=aime--aime_2026&comp=aime--aime_2026","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - APEX 2025":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=apex--apex_2025&amp%3Bcomp=apex--apex_2025&comp=apex--apex_2025","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"sources":[{"name":"MathArena - APEX 2025","url":"https://matharena.ai/?view=problem&amp%3Bcompetition=apex--apex_2025&amp%3Bcomp=apex--apex_2025&comp=apex--apex_2025","metric":"Accuracy (%)"},{"name":"MathArena Apex","url":"https://benchmarklist.com/benchmarks/matharena_apex/","metric":"Score (self-reported)"}]},"MathArena - APEX Shortlist 2025":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=apex--shortlist_2025&amp%3Bcomp=apex--shortlist_2025&comp=apex--shortlist_2025","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - ARXIV April":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=arxiv--april&amp%3Bcomp=arxiv--april&comp=arxiv--april","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - ARXIV February":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=arxiv--february&amp%3Bcomp=arxiv--february&comp=arxiv--february","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - ARXIV June":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=arxiv--june&amp%3Bcomp=arxiv--june&comp=arxiv--june","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - ARXIV March":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=arxiv--march&amp%3Bcomp=arxiv--march&comp=arxiv--march","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - ARXIVLEAN March":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=arxivlean--march&amp%3Bcomp=arxivlean--march&comp=arxivlean--march","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - ARXIV_FALSE April":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=arxiv_false--april&amp%3Bcomp=arxiv_false--april&comp=arxiv_false--april","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - ARXIV_FALSE February":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=arxiv_false--february&amp%3Bcomp=arxiv_false--february&comp=arxiv_false--february","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - ARXIV_FALSE June":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=arxiv_false--june&amp%3Bcomp=arxiv_false--june&comp=arxiv_false--june","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - ARXIV_FALSE March":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=arxiv_false--march&amp%3Bcomp=arxiv_false--march&comp=arxiv_false--march","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - ArXiv Math Dec 2025":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=arxiv--december&amp%3Bcomp=arxiv--december&comp=arxiv--december","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - ArXiv Math Jan 2026":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=arxiv--january&amp%3Bcomp=arxiv--january&comp=arxiv--january","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - BRUMO 2025":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=brumo--brumo_2025&amp%3Bcomp=brumo--brumo_2025&comp=brumo--brumo_2025","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - CMIMC 2025":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=cmimc--cmimc_2025&amp%3Bcomp=cmimc--cmimc_2025&comp=cmimc--cmimc_2025","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - HMMT Feb 2025":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=hmmt--hmmt_feb_2025&amp%3Bcomp=hmmt--hmmt_feb_2025&comp=hmmt--hmmt_feb_2025","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"sources":[{"name":"MathArena - HMMT Feb 2025","url":"https://matharena.ai/?view=problem&amp%3Bcompetition=hmmt--hmmt_feb_2025&amp%3Bcomp=hmmt--hmmt_feb_2025&comp=hmmt--hmmt_feb_2025","metric":"Accuracy (%)"},{"name":"HMMT 2025","url":"https://benchmarklist.com/benchmarks/hmmt_2025/","metric":"Score (self-reported)"}]},"MathArena - HMMT Feb 2026":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=hmmt--hmmt_feb_2026&amp%3Bcomp=hmmt--hmmt_feb_2026&comp=hmmt--hmmt_feb_2026","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - HMMT Nov 2025":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=hmmt--hmmt_nov_2025&amp%3Bcomp=hmmt--hmmt_nov_2025&comp=hmmt--hmmt_nov_2025","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - IMC 2025":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=imc--imc_2025&amp%3Bcomp=imc--imc_2025&comp=imc--imc_2025","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - IMO 2025":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=imo--imo_2025&amp%3Bcomp=imo--imo_2025&comp=imo--imo_2025","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - IMProofBench Final Answers":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=improofbench--final_answer&amp%3Bcomp=improofbench--final_answer&comp=improofbench--final_answer","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - IMProofBench Proofs":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=improofbench--proofs&amp%3Bcomp=improofbench--proofs&comp=improofbench--proofs","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - Kangaroo 2025 Levels 1-2":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=kangaroo--kangaroo_2025_1-2&amp%3Bcomp=kangaroo--kangaroo_2025_1-2&comp=kangaroo--kangaroo_2025_1-2","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - Kangaroo 2025 Levels 11-12":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=kangaroo--kangaroo_2025_11-12&amp%3Bcomp=kangaroo--kangaroo_2025_11-12&comp=kangaroo--kangaroo_2025_11-12","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - Kangaroo 2025 Levels 3-4":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=kangaroo--kangaroo_2025_3-4&amp%3Bcomp=kangaroo--kangaroo_2025_3-4&comp=kangaroo--kangaroo_2025_3-4","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - Kangaroo 2025 Levels 5-6":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=kangaroo--kangaroo_2025_5-6&amp%3Bcomp=kangaroo--kangaroo_2025_5-6&comp=kangaroo--kangaroo_2025_5-6","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - Kangaroo 2025 Levels 7-8":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=kangaroo--kangaroo_2025_7-8&amp%3Bcomp=kangaroo--kangaroo_2025_7-8&comp=kangaroo--kangaroo_2025_7-8","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - Kangaroo 2025 Levels 9-10":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=kangaroo--kangaroo_2025_9-10&amp%3Bcomp=kangaroo--kangaroo_2025_9-10&comp=kangaroo--kangaroo_2025_9-10","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - Project Euler 943-970":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=euler--euler&amp%3Bcomp=euler--euler&comp=euler--euler","metric":"Accuracy (%, direct Project Euler problems 943-970)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - Project Euler 971-984":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=euler--euler&amp%3Bcomp=euler--euler&comp=euler--euler","metric":"Accuracy (%, direct Project Euler problems 971-984)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - Project Euler 985-988":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=euler--euler&amp%3Bcomp=euler--euler&comp=euler--euler","metric":"Accuracy (%, direct Project Euler problems 985-988)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - Putnam 2025":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=putnam--putnam_2025&amp%3Bcomp=putnam--putnam_2025&comp=putnam--putnam_2025","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - SMT 2025":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=smt--smt_2025&amp%3Bcomp=smt--smt_2025&comp=smt--smt_2025","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - USAMO 2025":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=usamo--usamo_2025&amp%3Bcomp=usamo--usamo_2025&comp=usamo--usamo_2025","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena - Usamo 2026":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=usamo--usamo_2026&amp%3Bcomp=usamo--usamo_2026&comp=usamo--usamo_2026","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena Arxiv":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=arxiv--may&amp%3Bcomp=arxiv--may&comp=arxiv--may","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathArena Arxiv False":{"url":"https://matharena.ai/?view=problem&amp%3Bcompetition=arxiv_false--may&amp%3Bcomp=arxiv_false--may&comp=arxiv_false--may","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathBench":{"url":"https://github.com/open-compass/MathBench","metric":"Average Score","theoretical_min":0.0,"theoretical_max":100.0},"MathConstraint":{"url":"https://benchmarklist.com/benchmarks/mathconstraint_automated_generation_of_verified_combinatorial_reasoning_instances_for_llms/","metric":"Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MathSpatial":{"url":"https://shuolucs.github.io/MathSpatial/","metric":"Overall Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathTutorBench":{"url":"https://benchmarklist.com/benchmarks/mathtutorbench/","metric":"Overall Average (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MathVision":{"url":"https://mathllm.github.io/mathvision/#leaderboard","metric":"Overall Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MathVista":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"sources":[{"name":"MathVista","url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)"},{"name":"Open LMM Reasoning - MathVista","url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)"},{"name":"OpenCompass Multimodal - MathVista","url":"https://rank.opencompass.org.cn","metric":"Score (%)"}]},"MechVQA":{"url":"https://benchmarklist.com/benchmarks/mechvqa_benchmarking_and_enhancing_multimodal_llms_on_comprehensive_mechanical_drawing_understanding/","metric":"Total (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MedAgentBench":{"url":"https://benchmarklist.com/benchmarks/medagentbench/","metric":"Overall SR (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MedAgentGym":{"url":"https://benchmarklist.com/benchmarks/medagentgym/","metric":"Average Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MedBench v5":{"url":"https://benchmarklist.com/benchmarks/medbench_v5/","metric":"CCR-Agent (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MedBrowseComp 121":{"url":"https://github.com/shan23chen/MedBrowseComp","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MedBrowseComp 50":{"url":"https://github.com/shan23chen/MedBrowseComp","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MedCTA":{"url":"https://benchmarklist.com/benchmarks/medcta/","metric":"Outcome Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MedCode":{"url":"https://benchmarklist.com/benchmarks/vals_medcode/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MedFrameQA":{"url":"https://benchmarklist.com/benchmarks/medframeqa/","metric":"Average Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MedGUIDE":{"url":"https://benchmarklist.com/benchmarks/medguide/","metric":"Weighted Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MedHELM":{"url":"https://benchmarklist.com/benchmarks/medhelm/","metric":"Mean Win Rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MedLayXPlain":{"url":"https://benchmarklist.com/benchmarks/medlayxplain/","metric":"S (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MedS-Bench - Clinical Outcome Prediction":{"url":"https://henrychur.github.io/MedS-Bench/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"MedS-Bench - Concept Explanation":{"url":"https://henrychur.github.io/MedS-Bench/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"MedS-Bench - Diagnosis":{"url":"https://henrychur.github.io/MedS-Bench/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"MedS-Bench - Information Extraction":{"url":"https://henrychur.github.io/MedS-Bench/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"MedS-Bench - MCQA":{"url":"https://henrychur.github.io/MedS-Bench/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"MedS-Bench - Named Entity Recognition":{"url":"https://henrychur.github.io/MedS-Bench/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"MedS-Bench - Text Classification":{"url":"https://henrychur.github.io/MedS-Bench/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"MedS-Bench - Text Summarization":{"url":"https://henrychur.github.io/MedS-Bench/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"MedS-Bench - Treatment Planning":{"url":"https://henrychur.github.io/MedS-Bench/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"MedScribe":{"url":"https://benchmarklist.com/benchmarks/vals_medscribe/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MedXpertQA":{"url":"https://benchmarklist.com/benchmarks/medxpertqa/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Medical Chronology LLM Benchmark":{"url":"https://huggingface.co/datasets/Superinsight/medical-chronology-benchmark","metric":"Composite Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medical Information Response Audit (MIRA)":{"url":"https://benchmarklist.com/benchmarks/medical_information_response_audit_mira/","metric":"Underinformative simplification (D3) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Medical MM Leaderboard - Average":{"url":"https://huggingface.co/spaces/wangrongsheng/RS-Medical-LLM-Leaderboard","metric":"Average Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medical MM Leaderboard - MMMU-Med":{"url":"https://huggingface.co/spaces/wangrongsheng/RS-Medical-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medical MM Leaderboard - MedXQA":{"url":"https://huggingface.co/spaces/wangrongsheng/RS-Medical-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medical MM Leaderboard - OMVQA":{"url":"https://huggingface.co/spaces/wangrongsheng/RS-Medical-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medical MM Leaderboard - PMC-VQA":{"url":"https://huggingface.co/spaces/wangrongsheng/RS-Medical-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medical MM Leaderboard - PathVQA":{"url":"https://huggingface.co/spaces/wangrongsheng/RS-Medical-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medical MM Leaderboard - SLAKE":{"url":"https://huggingface.co/spaces/wangrongsheng/RS-Medical-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medical MM Leaderboard - VQA-RAD":{"url":"https://huggingface.co/spaces/wangrongsheng/RS-Medical-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medical STT Benchmark":{"url":"https://github.com/Omi-Health/medical-STT-eval","metric":"WER (lower is better)","theoretical_min":0.0,"theoretical_max":null},"Medmarks - AgentClinic":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - CareQA EN":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - CareQA Open":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - HEAD-QA v2":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - HealthBench":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - HealthBench Consensus":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - HealthBench Hard":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - LongHealth Task 1":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - LongHealth Task 2":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - M-ARC":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MEDEC":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MMLU-Pro Health":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MTSamples Procedures":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MTSamples Replicate":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - Med-HALT Reasoning FCT":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - Med-HALT Reasoning NOTA":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MedCalc-Bench":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MedCaseReasoning":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MedConceptsQA Easy":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MedConceptsQA Hard":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MedConceptsQA Medium":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MedDialog":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MedExQA":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MedHallu Easy":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MedHallu Hard":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MedHallu Medium":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MedMCQA":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MedQA":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MedR-Bench 1-Turn":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MedR-Bench Free-Turn":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MedR-Bench Oracle":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MedXpertQA Reasoning":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MedXpertQA Understanding":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - Medbullets OP4":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - Medbullets OP5":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MedicationQA":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - MetaMedQA":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - PubHealthBench Freeform":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - PubHealthBench Reviewed":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - PubMedQA":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - SCTpublic":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - SuperGPQA Medicine Easy":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Medmarks - SuperGPQA Medicine Hard":{"url":"https://medmarks.ai/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"MemPoison - Clean Task":{"url":"https://arxiv.org/abs/2607.14651","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"MemPoison - Poisoned Task":{"url":"https://arxiv.org/abs/2607.14651","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Mercor APEX":{"url":"https://www.mercor.com/apex/apex-agents-leaderboard/","metric":"Best Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"Merge-Bench":{"url":"https://benchmarklist.com/benchmarks/merge_bench_resolve_merge_conflicts_with_large_language_models/","metric":"Equivalent text (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Metacognitive Probe":{"url":"https://benchmarklist.com/benchmarks/metacognitive_probe/","metric":"Mean(T1,T2,T4,T5) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MineBench":{"url":"https://minebench.ai/","metric":"Elo Rating","theoretical_min":null,"theoretical_max":null},"MineExplorer":{"url":"https://benchmarklist.com/benchmarks/mineexplorer/","metric":"Overall TSR (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Minerva-Ego":{"url":"https://benchmarklist.com/benchmarks/minerva_ego_spatiotemporal_hints_for_egocentric_video_understanding/","metric":"MCQ-Acc. % (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Mirror, Mirror on the Wall":{"url":"https://benchmarklist.com/benchmarks/mirror_mirror_on_the_wall_can_vlm_agents_tell_who_they_are_at_all/","metric":"TSA (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MirrorCode":{"url":"https://epoch.ai/MirrorCode","metric":"Solve@100 Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"MixEval":{"url":"https://mixeval.github.io/","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"MixRea":{"url":"https://benchmarklist.com/benchmarks/mixrea/","metric":"Implicit Information Ignorance Rate (I3R) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Mizan LLM Leaderboard":{"url":"https://huggingface.co/spaces/MCINext/mizan-llm-leaderboard","metric":"Average Score (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"MoNaCo":{"url":"https://tomerwolgithub.github.io/monaco/leaderboard.html","metric":"F1","theoretical_min":0.0,"theoretical_max":100.0},"Mobile-MMLU":{"url":"https://huggingface.co/spaces/MBZUAI-LLM/Mobile-MMLU-Challenge","metric":"Overall Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"MolDeTox":{"url":"https://benchmarklist.com/benchmarks/moldetox/","metric":"Acc.(%) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MolViBench":{"url":"https://benchmarklist.com/benchmarks/molvibench_evaluating_llms_on_molecular_vibe_coding/","metric":"Pass@1 rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Momento":{"url":"https://benchmarklist.com/benchmarks/momento_evaluating_persistent_memory_and_reasoning_with_multi_session_agentic_conversations/","metric":"Pass@k (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MonitoringBench":{"url":"https://benchmarklist.com/benchmarks/monitoringbench_semi_automated_red_teaming_for_agent_monitoring/","metric":"Baseline catch rate (%) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MuPPET":{"url":"https://benchmarklist.com/benchmarks/muppet_a_benchmark_for_contextual_privacy_of_llm_assistants_in_multi_party_conversations/","metric":"Multi-Party (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Multi-Docker-Eval":{"url":"https://github.com/Z2sJ4t/Multi-Docker-Eval","metric":"Resolved (%)","theoretical_min":0.0,"theoretical_max":100.0},"Multi-IF":{"url":"https://benchmarklist.com/benchmarks/multi_if/","metric":"Average (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Multi-domain Multi-modal Document Classificati":{"url":"https://benchmarklist.com/benchmarks/multi_domain_multi_modal_document_classification_benchmark_with_a_multi_level_taxonomy/","metric":"HF1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Multi-turn Debate (Lechmazur)":{"url":"https://github.com/lechmazur/debate","metric":"Bradley-Terry Rating","theoretical_min":null,"theoretical_max":null},"MultiEmo-Bench":{"url":"https://benchmarklist.com/benchmarks/multiemo_bench/","metric":"Cosine (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MultiNRC":{"url":"https://benchmarklist.com/benchmarks/multinrc/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Multilingual Factual Questions Net Score":{"url":"https://benchmarklist.com/benchmarks/multilingual_factual_questions_net_score/","metric":"Net score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Multilingual MMLU - Arabic":{"url":"https://huggingface.co/spaces/StarscreamDeceptions/Multilingual-MMLU-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Multilingual MMLU - Bengali":{"url":"https://huggingface.co/spaces/StarscreamDeceptions/Multilingual-MMLU-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Multilingual MMLU - Chinese":{"url":"https://huggingface.co/spaces/StarscreamDeceptions/Multilingual-MMLU-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Multilingual MMLU - French":{"url":"https://huggingface.co/spaces/StarscreamDeceptions/Multilingual-MMLU-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Multilingual MMLU - German":{"url":"https://huggingface.co/spaces/StarscreamDeceptions/Multilingual-MMLU-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Multilingual MMLU - Hindi":{"url":"https://huggingface.co/spaces/StarscreamDeceptions/Multilingual-MMLU-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Multilingual MMLU - Indonesian":{"url":"https://huggingface.co/spaces/StarscreamDeceptions/Multilingual-MMLU-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Multilingual MMLU - Italian":{"url":"https://huggingface.co/spaces/StarscreamDeceptions/Multilingual-MMLU-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Multilingual MMLU - Japanese":{"url":"https://huggingface.co/spaces/StarscreamDeceptions/Multilingual-MMLU-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Multilingual MMLU - Korean":{"url":"https://huggingface.co/spaces/StarscreamDeceptions/Multilingual-MMLU-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Multilingual MMLU - Portuguese":{"url":"https://huggingface.co/spaces/StarscreamDeceptions/Multilingual-MMLU-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Multilingual MMLU - Spanish":{"url":"https://huggingface.co/spaces/StarscreamDeceptions/Multilingual-MMLU-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Multilingual MMLU - Swahili":{"url":"https://huggingface.co/spaces/StarscreamDeceptions/Multilingual-MMLU-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Multilingual MMLU - Yoruba":{"url":"https://huggingface.co/spaces/StarscreamDeceptions/Multilingual-MMLU-Benchmark-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Multilingual MMLU Leaderboard":{"url":"https://huggingface.co/spaces/StarscreamDeceptions/Multilingual-MMLU-Benchmark-Leaderboard","metric":"Average Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Multivac Evaluation":{"url":"https://github.com/themultivac/multivac-evaluation","metric":"Avg Peer Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"MusTBENCH":{"url":"https://benchmarklist.com/benchmarks/mustbench/","metric":"Total Avg. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"MyPCBench":{"url":"https://mypcbench.com/","metric":"Perfect Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"NARRA-Gym for Evaluating Interactive Narrative":{"url":"https://benchmarklist.com/benchmarks/narra_gym_for_evaluating_interactive_narrative_agents/","metric":"StoryQ (self-reported)","theoretical_min":0.0,"theoretical_max":null},"NICE":{"url":"https://benchmarklist.com/benchmarks/nice_a_theory_grounded_diagnostic_benchmark_for_social_intelligence_of_llms/","metric":"Weighted All (self-reported)","theoretical_min":0.0,"theoretical_max":null},"NL2Repo":{"url":"https://benchmarklist.com/benchmarks/nl2repo/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"NL2Scratch":{"url":"https://benchmarklist.com/benchmarks/nl2scratch_an_executable_benchmark_and_evaluation_for_block_based_programming/","metric":"SAC_ref Avg (self-reported)","theoretical_min":0.0,"theoretical_max":null},"NPHardEval":{"url":"https://huggingface.co/spaces/NPHardEval/NPHardEval-leaderboard","metric":"Average Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"NPHardEval - EDP":{"url":"https://huggingface.co/spaces/NPHardEval/NPHardEval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"NPHardEval - GCP":{"url":"https://huggingface.co/spaces/NPHardEval/NPHardEval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"NPHardEval - GCP D":{"url":"https://huggingface.co/spaces/NPHardEval/NPHardEval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"NPHardEval - KSP":{"url":"https://huggingface.co/spaces/NPHardEval/NPHardEval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"NPHardEval - MSP":{"url":"https://huggingface.co/spaces/NPHardEval/NPHardEval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"NPHardEval - SAS":{"url":"https://huggingface.co/spaces/NPHardEval/NPHardEval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"NPHardEval - SPP":{"url":"https://huggingface.co/spaces/NPHardEval/NPHardEval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"NPHardEval - TSP":{"url":"https://huggingface.co/spaces/NPHardEval/NPHardEval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"NPHardEval - TSP D":{"url":"https://huggingface.co/spaces/NPHardEval/NPHardEval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"NRITYAM":{"url":"https://benchmarklist.com/benchmarks/nrityam/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"NRT-Bench":{"url":"https://benchmarklist.com/benchmarks/nrt_bench_benchmarking_multi_turn_red_teaming_of_llm_operator_agents_in_safety_critical_control_rooms/","metric":"Aggregate ASR (self-reported)","theoretical_min":0.0,"theoretical_max":null},"NVIDIA LLM Robustness - AGIEval Accuracy":{"url":"https://huggingface.co/datasets/nvidia/llm-robustness-leaderboard-evals","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"NVIDIA LLM Robustness - AGIEval Consistency":{"url":"https://huggingface.co/datasets/nvidia/llm-robustness-leaderboard-evals","metric":"Consistency (%)","theoretical_min":0.0,"theoretical_max":100.0},"NVIDIA LLM Robustness - MATH Accuracy":{"url":"https://huggingface.co/datasets/nvidia/llm-robustness-leaderboard-evals","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"NVIDIA LLM Robustness - MATH Consistency":{"url":"https://huggingface.co/datasets/nvidia/llm-robustness-leaderboard-evals","metric":"Consistency (%)","theoretical_min":0.0,"theoretical_max":100.0},"NVIDIA LLM Robustness - MMLU-Pro Accuracy":{"url":"https://huggingface.co/datasets/nvidia/llm-robustness-leaderboard-evals","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"NVIDIA LLM Robustness - MMLU-Pro Consistency":{"url":"https://huggingface.co/datasets/nvidia/llm-robustness-leaderboard-evals","metric":"Consistency (%)","theoretical_min":0.0,"theoretical_max":100.0},"NYT Connections Extended":{"url":"https://github.com/lechmazur/nyt-connections","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"NYT Connections Older Models":{"url":"https://github.com/lechmazur/nyt-connections","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"NYT Connections Original":{"url":"https://github.com/lechmazur/nyt-connections","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"NYU CTF Bench":{"url":"https://huggingface.co/spaces/FrontierAICybersecurity/Cybersecurity_leaderboard","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"NanoGPT-Bench":{"url":"https://www.intology.ai/blog/nanogpt-bench","metric":"% of Human Progress Recovered","theoretical_min":0.0,"theoretical_max":100.0},"NarrativeWorldBench":{"url":"https://benchmarklist.com/benchmarks/narrativeworldbench/","metric":"Plot-Beat F1 (h=50) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Natural Language to Mongosh":{"url":"https://huggingface.co/datasets/mongodb-eai/natural-language-to-mongosh","metric":"NeXMaNeR (%)","theoretical_min":0.0,"theoretical_max":100.0},"Natural2Code":{"url":"https://benchmarklist.com/benchmarks/natural2code/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"NatureBench":{"url":"https://frontisai.github.io/NatureBench/","metric":"Surpass-SOTA (%)","theoretical_min":0.0,"theoretical_max":100.0},"NeedleBench":{"url":"https://benchmarklist.com/benchmarks/needlebench/","metric":"Overall 128K score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"NeoEvalPlusN":{"url":"https://huggingface.co/datasets/ChuckMcSneed/NeoEvalPlusN_benchmark","metric":"Mean Score (0-6)","theoretical_min":0.0,"theoretical_max":6.0},"Nexus Function Calling - CVECPE":{"url":"https://huggingface.co/spaces/Nexusflow/Nexus_Function_Calling_Leaderboard","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"Nexus Function Calling - CVECPE Multi APIs":{"url":"https://huggingface.co/spaces/Nexusflow/Nexus_Function_Calling_Leaderboard","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"Nexus Function Calling - Climate":{"url":"https://huggingface.co/spaces/Nexusflow/Nexus_Function_Calling_Leaderboard","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"Nexus Function Calling - OTX":{"url":"https://huggingface.co/spaces/Nexusflow/Nexus_Function_Calling_Leaderboard","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"Nexus Function Calling - Overall":{"url":"https://huggingface.co/spaces/Nexusflow/Nexus_Function_Calling_Leaderboard","metric":"Mean Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"Nexus Function Calling - Places API":{"url":"https://huggingface.co/spaces/Nexusflow/Nexus_Function_Calling_Leaderboard","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"Nexus Function Calling - VT Multi Dependency":{"url":"https://huggingface.co/spaces/Nexusflow/Nexus_Function_Calling_Leaderboard","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"Nexus Function Calling - VT Multi Disconnected":{"url":"https://huggingface.co/spaces/Nexusflow/Nexus_Function_Calling_Leaderboard","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"Nexus Function Calling - VirusTotal":{"url":"https://huggingface.co/spaces/Nexusflow/Nexus_Function_Calling_Leaderboard","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"NoLiMa":{"url":"https://github.com/adobe-research/NoLiMa","metric":"Base Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"NonoBench":{"url":"https://www.nonobench.com/","metric":"Overall Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OCR-Robust":{"url":"https://benchmarklist.com/benchmarks/ocr_robust/","metric":"OCR1.0 Clean (self-reported)","theoretical_min":0.0,"theoretical_max":null},"OCRBench v2":{"url":"https://benchmarklist.com/benchmarks/ocrbench_v2/","metric":"Average (self-reported)","theoretical_min":0.0,"theoretical_max":null},"OCRBench-V2 (en)":{"url":"https://benchmarklist.com/benchmarks/ocrbench_v2_en/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"OCRBench-V2 (zh)":{"url":"https://benchmarklist.com/benchmarks/ocrbench_v2_zh/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"OJBench":{"url":"https://benchmarklist.com/benchmarks/ojbench/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"OPT-BENCH":{"url":"https://benchmarklist.com/benchmarks/opt_bench_evaluating_the_iterative_self_optimization_of_llm_agents_in_large_scale_search_spaces/","metric":"EG (20 steps) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"OR-Space":{"url":"https://benchmarklist.com/benchmarks/or_space_a_full_lifecycle_workspace_benchmark_for_industrial_optimization_agents/","metric":"Build Pass@1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ORAgentBench":{"url":"https://benchmarklist.com/benchmarks/oragentbench_can_llm_agents_solve_challenging_operations_research_tasks_end_to_end/","metric":"Pass Rate All (self-reported)","theoretical_min":0.0,"theoretical_max":null},"OSQ-Bench":{"url":"https://huggingface.co/spaces/Open-Style/OSQ-Leaderboard","metric":"Average","theoretical_min":0.0,"theoretical_max":100.0},"OSWorld":{"url":"https://os-world.github.io/","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"OSWorld (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"First-Attempt Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"OSWorld 2.0":{"url":"https://osworld-v2.xlang.ai/","metric":"Binary Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OSWorld 2.0 Partial":{"url":"https://osworld-v2.xlang.ai/","metric":"Partial Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OSWorld-MCP":{"url":"https://osworld-mcp.github.io/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OSWorld-Verified":{"url":"https://benchmarklist.com/benchmarks/osworld_verified/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"OSWorld-Verified (Fable/Mythos)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OTIS Mock AIME 2024-25":{"url":"https://epoch.ai/benchmarks/otis-mock-aime-2024-2025","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OVBench":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"AVG","theoretical_min":0.0,"theoretical_max":100.0},"OccuBench":{"url":"https://benchmarklist.com/benchmarks/occubench/","metric":"Completion rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"OckBench":{"url":"https://ockbench.github.io/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OdysseyBench":{"url":"https://benchmarklist.com/benchmarks/odysseybench/","metric":"Overall (Old) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Office Comprehension Bench":{"url":"https://arxiv.org/abs/2607.01245","metric":"Domain Q&A Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Office Comprehension Bench - File Fidelity Excel":{"url":"https://arxiv.org/abs/2607.01245","metric":"File Fidelity Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Office Comprehension Bench - File Fidelity PowerPoint":{"url":"https://arxiv.org/abs/2607.01245","metric":"File Fidelity Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Office Comprehension Bench - File Fidelity Word":{"url":"https://arxiv.org/abs/2607.01245","metric":"File Fidelity Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OfficeQA (Anthropic Internal Harness)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"OfficeQA Pro":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Correctness (%)","theoretical_min":0.0,"theoretical_max":100.0},"OfficeQA Pro (Anthropic Internal Harness)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"OlympiadBench":{"url":"https://github.com/OpenBMB/OlympiadBench","metric":"Avg Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OlympicArena":{"url":"https://gair-nlp.github.io/OlympicArena/#leaderboard","metric":"Overall Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Omi SOAP Note Safety Benchmark":{"url":"https://benchmarklist.com/benchmarks/omi_soap_note_safety/","metric":"Composite (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Omni-DeepSearch":{"url":"https://benchmarklist.com/benchmarks/omni_deepsearch/","metric":"Avg (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Omni-MATH":{"url":"https://omni-math.github.io/","metric":"Overall Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OmniAI OCR Leaderboard":{"url":"https://huggingface.co/datasets/getomni-ai/ocr-leaderboard","metric":"JSON Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OmniDocBench 1.5":{"url":"https://benchmarklist.com/benchmarks/omnidocbench_1_5/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"OmniGAIA - Art":{"url":"https://huggingface.co/spaces/RUC-NLPIR/OmniGAIA-Leaderboard","metric":"Art Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OmniGAIA - Finance":{"url":"https://huggingface.co/spaces/RUC-NLPIR/OmniGAIA-Leaderboard","metric":"Finance Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OmniGAIA - Food":{"url":"https://huggingface.co/spaces/RUC-NLPIR/OmniGAIA-Leaderboard","metric":"Food Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OmniGAIA - Geo":{"url":"https://huggingface.co/spaces/RUC-NLPIR/OmniGAIA-Leaderboard","metric":"Geography Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OmniGAIA - History":{"url":"https://huggingface.co/spaces/RUC-NLPIR/OmniGAIA-Leaderboard","metric":"History Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OmniGAIA - Movie":{"url":"https://huggingface.co/spaces/RUC-NLPIR/OmniGAIA-Leaderboard","metric":"Movie Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OmniGAIA - Overall":{"url":"https://huggingface.co/spaces/RUC-NLPIR/OmniGAIA-Leaderboard","metric":"Overall Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OmniGAIA - Science":{"url":"https://huggingface.co/spaces/RUC-NLPIR/OmniGAIA-Leaderboard","metric":"Science Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OmniGAIA - Sport":{"url":"https://huggingface.co/spaces/RUC-NLPIR/OmniGAIA-Leaderboard","metric":"Sport Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OmniGAIA - Tech":{"url":"https://huggingface.co/spaces/RUC-NLPIR/OmniGAIA-Leaderboard","metric":"Technology Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OmniGameArena - CueChase":{"url":"https://arxiv.org/html/2606.09826v1","metric":"Normalized game score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OmniGameArena - HandoffRun":{"url":"https://arxiv.org/html/2606.09826v1","metric":"Normalized game score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OmniGameArena - LastStand":{"url":"https://arxiv.org/html/2606.09826v1","metric":"Normalized game score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OmniGameArena - MonsterShoot":{"url":"https://arxiv.org/html/2606.09826v1","metric":"Normalized game score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OmniGameArena - ObstacleRun2D":{"url":"https://arxiv.org/html/2606.09826v1","metric":"Normalized game score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OmniGameArena - ObstacleRun3D":{"url":"https://arxiv.org/html/2606.09826v1","metric":"Normalized game score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OmniGameArena - SceneEscape":{"url":"https://arxiv.org/html/2606.09826v1","metric":"Normalized game score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OmniGameArena - SharedFloor":{"url":"https://arxiv.org/html/2606.09826v1","metric":"Normalized game score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OmniGameArena - SoloCraft":{"url":"https://arxiv.org/html/2606.09826v1","metric":"Normalized game score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OmniMatBench":{"url":"https://benchmarklist.com/benchmarks/omnimatbench/","metric":"Avg. Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"OmniToM":{"url":"https://benchmarklist.com/benchmarks/omnitom_benchmarking_theory_of_mind_in_llms_via_explicit_belief_modeling/","metric":"Stage 2 Overall Acc (self-reported)","theoretical_min":0.0,"theoretical_max":null},"OmniaBench":{"url":"https://scuuy.github.io/OmniaBench/","metric":"Overall Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"On Stable Long-Form Generation":{"url":"https://benchmarklist.com/benchmarks/on_stable_long_form_generation_benchmarking_and_mitigating_length_volatility/","metric":"LVC (self-reported)","theoretical_min":0.0,"theoretical_max":null},"On the Cultural Anachronism and Temporal Reaso":{"url":"https://benchmarklist.com/benchmarks/on_the_cultural_anachronism_and_temporal_reasoning_in_vision_language_models/","metric":"Overall Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Onyx Open LLM Leaderboard":{"url":"https://www.onyx.app/open-llm-leaderboard","metric":"MMLU-Pro","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - ALL":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Alghafa Average":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Alghafa Meta AR Dialects":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Alghafa Meta AR MSA":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Alghafa Multiple Choice Facts Truefalse Balanced Task":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Alghafa Multiple Choice Grounded Statement Soqal Task":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Alghafa Multiple Choice Grounded Statement Xglue Mlqa Task":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Alghafa Multiple Choice Rating Sentiment NO Neutral Task":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Alghafa Multiple Choice Rating Sentiment Task":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Alghafa Multiple Choice Sentiment Task":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic Exams":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Accounting (University)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Arabic Language (General)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Arabic Language (Grammar)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Arabic Language (High School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Arabic Language (Middle School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Arabic Language (Primary School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Average":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Biology (High School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Civics (High School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Civics (Middle School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Computer Science (High School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Computer Science (Middle School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Computer Science (Primary School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Computer Science (University)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Driving Test":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Economics (High School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Economics (Middle School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Economics (University)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU General Knowledge":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU General Knowledge (Middle School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU General Knowledge (Primary School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Geography (High School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Geography (Middle School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Geography (Primary School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Abstract Algebra":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Anatomy":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Astronomy":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Average":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Business Ethics":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Clinical Knowledge":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT College Biology":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT College Chemistry":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT College Computer Science":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT College Mathematics":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT College Medicine":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT College Physics":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Computer Security":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Conceptual Physics":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Econometrics":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Electrical Engineering":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Elementary Mathematics":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Formal Logic":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Global Facts":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT High School Biology":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT High School Chemistry":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT High School Computer Science":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT High School European History":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT High School Geography":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT High School Government AND Politics":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT High School Macroeconomics":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT High School Mathematics":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT High School Microeconomics":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT High School Physics":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT High School Psychology":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT High School Statistics":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT High School US History":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT High School World History":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Human Aging":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Human Sexuality":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT International LAW":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Jurisprudence":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Logical Fallacies":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Machine Learning":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Management":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Marketing":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Medical Genetics":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Miscellaneous":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Moral Disputes":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Moral Scenarios":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Nutrition":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Philosophy":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Prehistory":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Professional Accounting":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Professional LAW":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Professional Medicine":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Professional Psychology":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Public Relations":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Security Studies":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Sociology":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT US Foreign Policy":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT Virology":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU HT World Religions":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU History (High School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU History (Middle School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU History (Primary School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Islamic Studies":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Islamic Studies (High School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Islamic Studies (Middle School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Islamic Studies (Primary School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU LAW (Professional)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Management (University)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Math (Primary School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Natural Science (Middle School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Natural Science (Primary School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Philosophy (High School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Physics (High School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Political Science (University)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Social Science (Middle School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Arabic MMLU Social Science (Primary School)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Aratrust Average":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Aratrust Ethics":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Aratrust Illegal":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Aratrust MentalHealth":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Aratrust Offensive":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Aratrust PhysicalHealth":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Aratrust Privacy":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Aratrust Trustfulness":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Aratrust Unfairness":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Madinah QA Arabic Language (General)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Madinah QA Arabic Language (Grammar)":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM - Madinah QA Average":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Arabic LLM Leaderboard":{"url":"https://huggingface.co/spaces/OALL/Open-Arabic-LLM-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Chinese LLM - ARC Challenge":{"url":"https://huggingface.co/spaces/BAAI/open_cn_llm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Chinese LLM - C-Eval Semantic":{"url":"https://huggingface.co/spaces/BAAI/open_cn_llm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Chinese LLM - CMMLU":{"url":"https://huggingface.co/spaces/BAAI/open_cn_llm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Chinese LLM - GSM8K":{"url":"https://huggingface.co/spaces/BAAI/open_cn_llm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Chinese LLM - HellaSwag":{"url":"https://huggingface.co/spaces/BAAI/open_cn_llm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Chinese LLM - TruthfulQA MC":{"url":"https://huggingface.co/spaces/BAAI/open_cn_llm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Chinese LLM - WinoGrande":{"url":"https://huggingface.co/spaces/BAAI/open_cn_llm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Chinese LLM Leaderboard":{"url":"https://huggingface.co/spaces/BAAI/open_cn_llm_leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open CoT - LSAT Analytical Reasoning":{"url":"https://huggingface.co/spaces/logikon/open_cot_leaderboard","metric":"CoT Gain (%)","theoretical_min":null,"theoretical_max":null},"Open CoT - LSAT Logical Reasoning":{"url":"https://huggingface.co/spaces/logikon/open_cot_leaderboard","metric":"CoT Gain (%)","theoretical_min":null,"theoretical_max":null},"Open CoT - LSAT Reading Comprehension":{"url":"https://huggingface.co/spaces/logikon/open_cot_leaderboard","metric":"CoT Gain (%)","theoretical_min":null,"theoretical_max":null},"Open CoT - LogiQA":{"url":"https://huggingface.co/spaces/logikon/open_cot_leaderboard","metric":"CoT Gain (%)","theoretical_min":null,"theoretical_max":null},"Open CoT - LogiQA 2":{"url":"https://huggingface.co/spaces/logikon/open_cot_leaderboard","metric":"CoT Gain (%)","theoretical_min":null,"theoretical_max":null},"Open CoT Leaderboard":{"url":"https://huggingface.co/spaces/logikon/open_cot_leaderboard","metric":"Average CoT Gain (%)","theoretical_min":null,"theoretical_max":null},"Open FinLLM Leaderboard":{"url":"https://benchmarklist.com/benchmarks/open_finllm_leaderboard/","metric":"Average normalized score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Open FinLLM Reasoning":{"url":"https://huggingface.co/spaces/TheFinAI/open-finllm-reasoning-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open FinLLM Reasoning - DM-Complong":{"url":"https://huggingface.co/spaces/TheFinAI/open-finllm-reasoning-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open FinLLM Reasoning - DM-Simplong":{"url":"https://huggingface.co/spaces/TheFinAI/open-finllm-reasoning-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open FinLLM Reasoning - FinQA":{"url":"https://huggingface.co/spaces/TheFinAI/open-finllm-reasoning-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open FinLLM Reasoning - XBRL-Math":{"url":"https://huggingface.co/spaces/TheFinAI/open-finllm-reasoning-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Financial LLM - BloombergGPT":{"url":"https://huggingface.co/spaces/finosfoundation/Open-Financial-LLM-Leaderboard","metric":"Average Score","theoretical_min":0.0,"theoretical_max":100.0},"Open Italian LLM - MMLU-Pro (IT)":{"url":"https://huggingface.co/spaces/mii-llm/open_ita_llm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - AIO Char F1":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - ALT E TO J Bert Score JA F1":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - ALT E TO J Bleu JA":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - ALT E TO J Comet Wmt22":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - ALT J TO E Bert Score EN F1":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - ALT J TO E Bleu EN":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - ALT J TO E Comet Wmt22":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - CG":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Commonsensemoralja Exact Match":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - EL":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - FA":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - HE":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Jamp Exact Match":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Janli Exact Match":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Jcommonsenseqa Exact Match":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Jemhopqa Char F1":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Jmmlu Exact Match":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Jnli Exact Match":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Jsem Exact Match":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Jsick Exact Match":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Jsts Pearson":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":null,"theoretical_max":null},"Open Japanese LLM - Jsts Spearman":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":null,"theoretical_max":null},"Open Japanese LLM - Kuci Exact Match":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - MC":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - MMLU EN Exact Match":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - MR":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - MT":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Mbpp Pylint Check":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - NLI":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Niilc Char F1":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - QA":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - RC":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - SUM":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Wiki Coreference SET F1":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Wiki Dependency SET F1":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Wiki NER SET F1":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Wiki PAS SET F1":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Wiki Reading Char F1":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Wikicorpus E TO J Bert Score JA F1":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Wikicorpus E TO J Bleu JA":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Wikicorpus E TO J Comet Wmt22":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Wikicorpus J TO E Bert Score EN F1":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Wikicorpus J TO E Bleu EN":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Wikicorpus J TO E Comet Wmt22":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Xlsum JA Bert Score JA F1":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Xlsum JA Bleu JA":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Xlsum JA Rouge1":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Xlsum JA Rouge2":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM - Xlsum JA RougeLsum":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Japanese LLM Leaderboard":{"url":"https://huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Korean LLM Leaderboard":{"url":"https://huggingface.co/spaces/upstage/open-ko-llm-leaderboard","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":null},"Open LLM Leaderboard - BBH":{"url":"https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"Open LLM Leaderboard - GPQA":{"url":"https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"Open LLM Leaderboard - IFEval":{"url":"https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"Open LLM Leaderboard - MATH Level 5":{"url":"https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"Open LLM Leaderboard - MMLU-Pro":{"url":"https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"Open LLM Leaderboard - MuSR":{"url":"https://huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Level-elementary school":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Level-elementary school; Avg":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Level-high school":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Level-high school; Avg":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Level-undergraduate":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Level-undergraduate; Avg":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Overall; Avg":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Subject-algebra":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Subject-algebra; Avg":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Subject-analytic geometry":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Subject-analytic geometry; Avg":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Subject-arithmetic":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Subject-arithmetic; Avg":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Subject-graph theory":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Subject-graph theory; Avg":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Subject-plane geometry":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Subject-plane geometry; Avg":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Subject-puzzle test":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Subject-puzzle test; Avg":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Subject-scientific figure":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Subject-scientific figure; Avg":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Subject-solid geometry":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Subject-solid geometry; Avg":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Subject-statistics":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - DynaMath - Subject-statistics; Avg":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - LogicVista":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - LogicVista - deductive":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - LogicVista - inductive":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - LogicVista - mechanical":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - LogicVista - numerical":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - LogicVista - spatial":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MMMath":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MMMath - Difficulty-easy":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MMMath - Difficulty-hard":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MMMath - Difficulty-medium":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MMMath - Knowledge-L1-Functions":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MMMath - Knowledge-L1-Properties of Shapes":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MMMath - Knowledge-L1-Transformations of Shapes":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MMMath - Knowledge-L2-Acute Angle Trigonometric Functions":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MMMath - Knowledge-L2-Circle":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MMMath - Knowledge-L2-Intersecting and Parallel Lines":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MMMath - Knowledge-L2-Inverse Proportional Function":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MMMath - Knowledge-L2-Linear Function":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MMMath - Knowledge-L2-Quadratic Function":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MMMath - Knowledge-L2-Quadrilateral":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MMMath - Knowledge-L2-Similarity of Shapes":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MMMath - Knowledge-L2-Symmetry of Shapes":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MMMath - Knowledge-L2-Triangle":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MMMath - Year-eight":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MMMath - Year-nine":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MMMath - Year-seven":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVerse":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVerse - Analytic":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVerse - Angle":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVerse - Applied":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVerse - Area":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVerse - Coordinate":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVerse - Expression":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVerse - Functions":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVerse - Length":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVerse - Plane Geometry":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVerse - Property":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVerse - Solid Geometry":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVerse - Volume":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVision":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVision - algebra":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVision - analytic geometry":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVision - arithmetic":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVision - combinatorial geometry":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVision - combinatorics":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVision - counting":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVision - descriptive geometry":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVision - graph theory":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVision - logic":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVision - metric geometry - angle":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVision - metric geometry - area":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVision - metric geometry - length":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVision - solid geometry":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVision - statistics":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVision - topology":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVision - transformation geometry":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVista - ALG":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVista - ARI":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVista - FQA":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVista - GEO":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVista - GPS":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVista - LOG":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVista - MWP":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVista - NUM":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVista - SCI":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVista - STA":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVista - TQA":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - MathVista - VQA":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - OlympiadBench":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - OlympiadBench - CN":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - OlympiadBench - EN":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - OlympiadBench - Geometry":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - OlympiadBench - Plane Geometry":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - OlympiadBench - Probability and Statistics":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - OlympiadBench - Solid Geometry":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - OlympiadBench - Trigonometric Functions":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - Angles and Length":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - Basic Transformations of Figures":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - Calculation of Plane Figures":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - Calculation of Solid Figures":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - CompleteMastery (Loose)":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - CompleteMastery (Strict)":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - Correspondence of Coordinates and Positions":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - Cutting and Combining of Figures":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - Direction":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - InadequateGeneralization (Loose)":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - InsufficientKnowledge (Loose)":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - One-step(S1)":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - Position":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - RoteMemorization (Loose)":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - RoteMemorization (Strict)":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - Route Map":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - Score (Loose)":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - Three-step(S3)":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - Two-step(S2)":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - Understanding and Conversion of Units":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - Understanding of Plane Figures":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open LMM Reasoning - WeMath - Understanding of Solid Figures":{"url":"https://huggingface.co/spaces/opencompass/Open_LMM_Reasoning_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Medical LLM":{"url":"https://huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","metric":"Average Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Medical LLM - MMLU Anatomy":{"url":"https://huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":25},"Open Medical LLM - MMLU Clinical Knowledge":{"url":"https://huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":25},"Open Medical LLM - MMLU College Biology":{"url":"https://huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":25},"Open Medical LLM - MMLU College Medicine":{"url":"https://huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":25},"Open Medical LLM - MMLU Medical Genetics":{"url":"https://huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":25},"Open Medical LLM - MMLU Professional Medicine":{"url":"https://huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":25},"Open Medical LLM - MedMCQA":{"url":"https://huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Medical LLM - MedQA (USMLE)":{"url":"https://huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Medical LLM - PubMedQA":{"url":"https://huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Multilingual Reasoning (13B) - MGSM":{"url":"https://huggingface.co/spaces/kevinpro/Open-Multilingual-Reasoning-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Multilingual Reasoning (13B) - MNumGLUESub":{"url":"https://huggingface.co/spaces/kevinpro/Open-Multilingual-Reasoning-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Multilingual Reasoning (13B) - MSVAMP":{"url":"https://huggingface.co/spaces/kevinpro/Open-Multilingual-Reasoning-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Multilingual Reasoning (7B) - MGSM":{"url":"https://huggingface.co/spaces/kevinpro/Open-Multilingual-Reasoning-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Multilingual Reasoning (7B) - MNumGLUESub":{"url":"https://huggingface.co/spaces/kevinpro/Open-Multilingual-Reasoning-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Multilingual Reasoning (7B) - MSVAMP":{"url":"https://huggingface.co/spaces/kevinpro/Open-Multilingual-Reasoning-Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open PL LLM - Generative":{"url":"https://huggingface.co/spaces/speakleash/open_pl_llm_leaderboard","metric":"Average Generative Score (%)","theoretical_min":null,"theoretical_max":100.0},"Open PL LLM - Multiple Choice":{"url":"https://huggingface.co/spaces/speakleash/open_pl_llm_leaderboard","metric":"Average Multiple-Choice Score (%)","theoretical_min":null,"theoretical_max":100.0},"Open PL LLM - RAG":{"url":"https://huggingface.co/spaces/speakleash/open_pl_llm_leaderboard","metric":"Average RAG Score (%)","theoretical_min":null,"theoretical_max":100.0},"Open PL LLM Leaderboard":{"url":"https://huggingface.co/spaces/speakleash/open_pl_llm_leaderboard","metric":"Average Score (%)","theoretical_min":null,"theoretical_max":100.0},"Open Persian LLM - ARC Challenge":{"url":"https://huggingface.co/spaces/PartAI/open-persian-llm-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Persian LLM - ARC Easy":{"url":"https://huggingface.co/spaces/PartAI/open-persian-llm-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Persian LLM - AUT Multiple Choice Persian":{"url":"https://huggingface.co/spaces/PartAI/open-persian-llm-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Persian LLM - MMLU Pro":{"url":"https://huggingface.co/spaces/PartAI/open-persian-llm-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Persian LLM - Part Multiple Choice":{"url":"https://huggingface.co/spaces/PartAI/open-persian-llm-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Persian LLM Leaderboard":{"url":"https://huggingface.co/spaces/PartAI/open-persian-llm-leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Portuguese LLM - ASSIN2 STS":{"url":"https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard","metric":"Pearson Correlation (×100)","theoretical_min":null,"theoretical_max":null},"Open Portuguese LLM - Hate Speech":{"url":"https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard","metric":"F1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Portuguese LLM - HateBR Offensive":{"url":"https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard","metric":"F1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Portuguese LLM - TweetSentBR":{"url":"https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard","metric":"F1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Portuguese LLM Leaderboard":{"url":"https://huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Open Universal Arabic ASR Leaderboard":{"url":"https://huggingface.co/spaces/elmresearchcenter/open_universal_arabic_asr_leaderboard","metric":"Average WER (lower is better)","theoretical_min":0.0,"theoretical_max":null},"Open Universal Arabic ASR Leaderboard - CER":{"url":"https://huggingface.co/spaces/elmresearchcenter/open_universal_arabic_asr_leaderboard","metric":"Average CER (lower is better)","theoretical_min":0.0,"theoretical_max":null},"Open-R1 Eval Leaderboard":{"url":"https://huggingface.co/spaces/open-r1/open-r1-eval-leaderboard","metric":"Average Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI CTF (Professional)":{"url":"https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf","metric":"pass@12 (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI ChatGPT Pro - AIME 2024":{"url":"https://openai.com/index/introducing-chatgpt-pro/","metric":"pass@1 accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI ChatGPT Pro - AIME 2024 4/4 Reliability":{"url":"https://openai.com/index/introducing-chatgpt-pro/","metric":"4/4 reliability (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI ChatGPT Pro - Codeforces":{"url":"https://openai.com/index/introducing-chatgpt-pro/","metric":"pass@1 percentile","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI ChatGPT Pro - Codeforces 4/4 Reliability":{"url":"https://openai.com/index/introducing-chatgpt-pro/","metric":"4/4 reliability percentile","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI ChatGPT Pro - GPQA Diamond":{"url":"https://openai.com/index/introducing-chatgpt-pro/","metric":"pass@1 accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI ChatGPT Pro - GPQA Diamond 4/4 Reliability":{"url":"https://openai.com/index/introducing-chatgpt-pro/","metric":"4/4 reliability (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI Cyber Ranges":{"url":"https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf","metric":"Combined Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5 System Card - BBQ Ambiguous":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Accuracy","theoretical_min":0.0,"theoretical_max":1.0},"OpenAI GPT-5 System Card - BBQ Disambiguated":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Accuracy","theoretical_min":0.0,"theoretical_max":1.0},"OpenAI GPT-5 System Card - HealthBench":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5 System Card - HealthBench Consensus":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5 System Card - HealthBench Hard":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5 System Card - Human Pathogen Capabilities Test":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Mean Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5 System Card - MMLU Language Arabic":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Accuracy","theoretical_min":0.0,"theoretical_max":1.0},"OpenAI GPT-5 System Card - MMLU Language Bengali":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Accuracy","theoretical_min":0.0,"theoretical_max":1.0},"OpenAI GPT-5 System Card - MMLU Language Chinese":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Accuracy","theoretical_min":0.0,"theoretical_max":1.0},"OpenAI GPT-5 System Card - MMLU Language French":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Accuracy","theoretical_min":0.0,"theoretical_max":1.0},"OpenAI GPT-5 System Card - MMLU Language German":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Accuracy","theoretical_min":0.0,"theoretical_max":1.0},"OpenAI GPT-5 System Card - MMLU Language Hindi":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Accuracy","theoretical_min":0.0,"theoretical_max":1.0},"OpenAI GPT-5 System Card - MMLU Language Indonesian":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Accuracy","theoretical_min":0.0,"theoretical_max":1.0},"OpenAI GPT-5 System Card - MMLU Language Italian":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Accuracy","theoretical_min":0.0,"theoretical_max":1.0},"OpenAI GPT-5 System Card - MMLU Language Japanese":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Accuracy","theoretical_min":0.0,"theoretical_max":1.0},"OpenAI GPT-5 System Card - MMLU Language Korean":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Accuracy","theoretical_min":0.0,"theoretical_max":1.0},"OpenAI GPT-5 System Card - MMLU Language Portuguese":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Accuracy","theoretical_min":0.0,"theoretical_max":1.0},"OpenAI GPT-5 System Card - MMLU Language Spanish":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Accuracy","theoretical_min":0.0,"theoretical_max":1.0},"OpenAI GPT-5 System Card - MMLU Language Swahili":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Accuracy","theoretical_min":0.0,"theoretical_max":1.0},"OpenAI GPT-5 System Card - MMLU Language Yoruba":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Accuracy","theoretical_min":0.0,"theoretical_max":1.0},"OpenAI GPT-5 System Card - Molecular Biology Capabilities Test":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Mean Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5 System Card - Virology Capabilities Test":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Mean Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5 System Card - Virology Capabilities Test Text-Only":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Mean Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5 System Card - World Class Biology":{"url":"https://cdn.openai.com/gpt-5-system-card.pdf","metric":"Mean Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.4 Launch - ARC-AGI-1 (Verified)":{"url":"https://openai.com/index/introducing-gpt-5-4/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.4 Launch - ARC-AGI-2 (Verified)":{"url":"https://openai.com/index/introducing-gpt-5-4/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.4 Launch - BrowseComp":{"url":"https://openai.com/index/introducing-gpt-5-4/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.4 Launch - FinanceAgent v1.1":{"url":"https://openai.com/index/introducing-gpt-5-4/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.4 Launch - Frontier Science Research":{"url":"https://openai.com/index/introducing-gpt-5-4/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.4 Launch - FrontierMath Tier 1-3":{"url":"https://openai.com/index/introducing-gpt-5-4/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.4 Launch - FrontierMath Tier 4":{"url":"https://openai.com/index/introducing-gpt-5-4/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.4 Launch - GDPval":{"url":"https://openai.com/index/introducing-gpt-5-4/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.4 Launch - GPQA Diamond":{"url":"https://openai.com/index/introducing-gpt-5-4/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.4 Launch - Humanity's Last Exam (no tools)":{"url":"https://openai.com/index/introducing-gpt-5-4/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.4 Launch - Humanity's Last Exam (with tools)":{"url":"https://openai.com/index/introducing-gpt-5-4/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.4 Launch - Investment Banking Modeling Tasks":{"url":"https://openai.com/index/introducing-gpt-5-4/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.5 Launch - ARC-AGI-1 (Verified)":{"url":"https://openai.com/index/introducing-gpt-5-5/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.5 Launch - ARC-AGI-2 (Verified)":{"url":"https://openai.com/index/introducing-gpt-5-5/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.5 Launch - BrowseComp":{"url":"https://openai.com/index/introducing-gpt-5-5/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.5 Launch - FinanceAgent v1.1":{"url":"https://openai.com/index/introducing-gpt-5-5/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.5 Launch - FrontierMath Tier 1-3":{"url":"https://openai.com/index/introducing-gpt-5-5/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.5 Launch - FrontierMath Tier 4":{"url":"https://openai.com/index/introducing-gpt-5-5/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.5 Launch - GDPval (wins or ties)":{"url":"https://openai.com/index/introducing-gpt-5-5/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.5 Launch - GPQA Diamond":{"url":"https://openai.com/index/introducing-gpt-5-5/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.5 Launch - GeneBench":{"url":"https://openai.com/index/introducing-gpt-5-5/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.5 Launch - Humanity's Last Exam (no tools)":{"url":"https://openai.com/index/introducing-gpt-5-5/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.5 Launch - Humanity's Last Exam (with tools)":{"url":"https://openai.com/index/introducing-gpt-5-5/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.5 Launch - Investment Banking Modeling Tasks":{"url":"https://openai.com/index/introducing-gpt-5-5/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.5 System Card - Biochemistry Knowledge Improvement":{"url":"https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf","metric":"reward@4 (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.5 System Card - DNA Sequence Design for TF Binding":{"url":"https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf","metric":"pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.5 System Card - Hard Negative Protein Binding Prediction":{"url":"https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf","metric":"pass@4 (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-5.5 System Card - Tacit Knowledge and Troubleshooting":{"url":"https://deploymentsafety.openai.com/gpt-5-5/gpt-5-5.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-Rosalind-5.5 System Card - Biorisk Knowledge":{"url":"https://deploymentsafety.openai.com/gpt-rosalind-5-5/gpt-rosalind-5-5.pdf","metric":"cons@32 (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-Rosalind-5.5 System Card - DNA Sequence Design for TF Binding":{"url":"https://deploymentsafety.openai.com/gpt-rosalind-5-5/gpt-rosalind-5-5.pdf","metric":"pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-Rosalind-5.5 System Card - Hard Negative Protein Binding Prediction":{"url":"https://deploymentsafety.openai.com/gpt-rosalind-5-5/gpt-rosalind-5-5.pdf","metric":"pass@4 (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-Rosalind-5.5 System Card - Multi-select Virology Troubleshooting":{"url":"https://deploymentsafety.openai.com/gpt-rosalind-5-5/gpt-rosalind-5-5.pdf","metric":"pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-Rosalind-5.5 System Card - ProtocolQA Open-Ended":{"url":"https://deploymentsafety.openai.com/gpt-rosalind-5-5/gpt-rosalind-5-5.pdf","metric":"pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI GPT-Rosalind-5.5 System Card - TroubleshootingBench":{"url":"https://deploymentsafety.openai.com/gpt-rosalind-5-5/gpt-rosalind-5-5.pdf","metric":"pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI HumanEval":{"url":"https://huggingface.co/datasets/evaleval/EEE_datastore","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenAI LifeSciBench":{"url":"https://openai.com/index/introducing-life-sci-bench/","metric":"Exact Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenBookQA":{"url":"https://epoch.ai/benchmarks/eci","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":25},"OpenClaw Arena Model Leaderboard":{"url":"https://benchmarklist.com/benchmarks/openclaw_arena_models/","metric":"Avg Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"OpenClawProBench":{"url":"https://suyoumo.github.io/bench/","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass":{"url":"https://rank.opencompass.org.cn","metric":"Average Score","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Agent - Multi-Turn":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Agent - Tool Use":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Code - Competition":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Code - Comprehensive":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Knowledge - Common Sense":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Knowledge - Engineering":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Knowledge - Humanities":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Knowledge - Science":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Knowledge - Social Science":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass LLM - Agent":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass LLM - Code":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass LLM - Knowledge":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass LLM - Language":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass LLM - Math":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass LLM - Reasoning":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Language - Chat":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Language - Creation":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Language - Dialogue":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Language - Instruction Following":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Language - NLP":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Math - College":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Math - Competition":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Multimodal - Average":{"url":"https://rank.opencompass.org.cn","metric":"Average Score","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Multimodal - MMBench":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Multimodal - MMMU":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Multimodal - OCRBench":{"url":"https://rank.opencompass.org.cn","metric":"Score (0-1000)","theoretical_min":0.0,"theoretical_max":1000.0},"OpenCompass Reasoning - Academic":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Reasoning - Common":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Research - AIME 2025":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Research - GPQA Diamond":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Research - HLE":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Research - IFEval":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Research - LiveCodeBench v6":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenCompass Research - MMLU-Pro":{"url":"https://rank.opencompass.org.cn","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenComputer":{"url":"https://benchmarklist.com/benchmarks/opencomputer_verifiable_software_worlds_for_computer_use_agents/","metric":"Success Rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"OpenEval - BBQ":{"url":"https://huggingface.co/datasets/human-centered-eval/OpenEval","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenEval - BoolQ":{"url":"https://huggingface.co/datasets/human-centered-eval/OpenEval","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenEval - CNN/DailyMail":{"url":"https://huggingface.co/datasets/human-centered-eval/OpenEval","metric":"ROUGE-L","theoretical_min":0.0,"theoretical_max":100.0},"OpenEval - GPQA CoT":{"url":"https://huggingface.co/datasets/human-centered-eval/OpenEval","metric":"CoT Correctness (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenEval - IFEval Strict":{"url":"https://huggingface.co/datasets/human-centered-eval/OpenEval","metric":"Strict Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenEval - IMDb":{"url":"https://huggingface.co/datasets/human-centered-eval/OpenEval","metric":"Exact Match (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenEval - MMLU-Pro CoT":{"url":"https://huggingface.co/datasets/human-centered-eval/OpenEval","metric":"CoT Correctness (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenEval - Omni-MATH":{"url":"https://huggingface.co/datasets/human-centered-eval/OpenEval","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenEval - XSum":{"url":"https://huggingface.co/datasets/human-centered-eval/OpenEval","metric":"ROUGE-L","theoretical_min":0.0,"theoretical_max":100.0},"OpenEvals - AIME 2026":{"url":"https://huggingface.co/spaces/OpenEvals/every-leaderboards","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenEvals - EvasionBench":{"url":"https://huggingface.co/spaces/OpenEvals/every-leaderboards","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenEvals - GPQA Diamond":{"url":"https://huggingface.co/spaces/OpenEvals/every-leaderboards","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenEvals - GSM8K":{"url":"https://huggingface.co/spaces/OpenEvals/every-leaderboards","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenEvals - HMMT February 2026":{"url":"https://huggingface.co/spaces/OpenEvals/every-leaderboards","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenEvals - Humanity's Last Exam":{"url":"https://huggingface.co/spaces/OpenEvals/every-leaderboards","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenEvals - MMLU-Pro":{"url":"https://huggingface.co/spaces/OpenEvals/every-leaderboards","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenEvals - SWE-bench Pro":{"url":"https://huggingface.co/spaces/OpenEvals/every-leaderboards","metric":"Resolved (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenEvals - SWE-bench Verified":{"url":"https://huggingface.co/spaces/OpenEvals/every-leaderboards","metric":"Resolved (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenEvals - Terminal-Bench 2.0":{"url":"https://huggingface.co/spaces/OpenEvals/every-leaderboards","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenExempt":{"url":"https://benchmarklist.com/benchmarks/openexempt/","metric":"Mean F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"OpenHuEval":{"url":"https://benchmarklist.com/benchmarks/openhueval/","metric":"Macro Average (computed) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"OpenRCA":{"url":"https://microsoft.github.io/OpenRCA/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenRCA Banking (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Root Cause Identification (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenRCA Market (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Root Cause Identification (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenRCA Overall (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Root Cause Identification (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenRCA Telecom (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Root Cause Identification (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenSkillEval":{"url":"https://benchmarklist.com/benchmarks/openskilleval_automatically_auditing_the_open_skill_ecosystem_for_llm_agents/","metric":"Overall avg. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"OpenVLM A-Bench":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM AI2D - Atom Structure":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM AI2D - Eclipses":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM AI2D - Faults and Earthquakes":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM AI2D - Food Chains and Webs":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM AI2D - Life Cycles":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM AI2D - Moon Phase and Equinox":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM AI2D - Parts of a Whole":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM AI2D - Parts of the Earth":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM AI2D - Photosynthesis and Respiration":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM AI2D - Rock Cycle":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM AI2D - Rock Strata":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM AI2D - Solar System":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM AI2D - Types Of":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM AI2D - Volcano":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM AI2D - Water, Carbon, and Nitrogen Cycle":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM COCO Captions":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"CIDEr","theoretical_min":0.0,"theoretical_max":null},"OpenVLM HallusionBench - aAcc":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM HallusionBench - fAcc":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM HallusionBench - qAcc":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM LLaVA-Bench - Complex":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":null},"OpenVLM LLaVA-Bench - Conv":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":null},"OpenVLM LLaVA-Bench - Detail":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":null},"OpenVLM MMBench V1.1 CN - AR":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - Action Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - Attribute Comparison":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - Attribute Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - CP":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - Celebrity Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - FP-C":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - FP-S":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - Function Reasoning":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - Future Prediction":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - Identity Reasoning":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - Image Emotion":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - Image Quality":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - Image Scene":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - Image Style":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - Image Topic":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - LR":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - Nature Relation":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - OCR":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - Object Localization":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - Overall":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - Physical Property Reasoning":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - Physical Relation":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - RR":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - Social Relation":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - Spatial Relationship":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 CN - Structuralized Imagetext Understanding":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - AR":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - Action Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - Attribute Comparison":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - Attribute Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - CP":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - Celebrity Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - FP-C":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - FP-S":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - Function Reasoning":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - Future Prediction":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - Identity Reasoning":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - Image Emotion":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - Image Quality":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - Image Scene":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - Image Style":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - Image Topic":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - LR":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - Nature Relation":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - OCR":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - Object Localization":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - Overall":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - Physical Property Reasoning":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - Physical Relation":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - RR":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - Social Relation":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - Spatial Relationship":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMBench V1.1 EN - Structuralized Imagetext Understanding":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MME":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Overall Score","theoretical_min":0.0,"theoretical_max":2800.0},"OpenVLM MME - Artwork":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":200.0},"OpenVLM MME - Celebrity":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":200.0},"OpenVLM MME - Code Reasoning":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":200.0},"OpenVLM MME - Cognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":800.0},"OpenVLM MME - Color":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":200.0},"OpenVLM MME - Commonsense Reasoning":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":200.0},"OpenVLM MME - Count":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":200.0},"OpenVLM MME - Existence":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":200.0},"OpenVLM MME - Landmark":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":200.0},"OpenVLM MME - Numerical Calculation":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":200.0},"OpenVLM MME - OCR":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":200.0},"OpenVLM MME - Perception":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":2000.0},"OpenVLM MME - Position":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":200.0},"OpenVLM MME - Posters":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":200.0},"OpenVLM MME - Scene":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":200.0},"OpenVLM MME - Text Translation":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":200.0},"OpenVLM MMMU - Accounting":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Agriculture":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Architecture and Engineering":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Art":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Art Theory":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Art and Design":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Basic Medical Science":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Biology":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Business":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Chemistry":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Clinical Medicine":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Computer Science":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Design":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Diagnostics and Laboratory Medicine":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Economics":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Electronics":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Energy and Power":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Finance":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Geography":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Health and Medicine":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - History":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Humanities and Social Science":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Literature":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Management":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Marketing":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Materials":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Math":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Mechanical Engineering":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Music":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Pharmacy":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Physics":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Psychology":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Public Health":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Science":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Sociology":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMMU - Tech and Engineering":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMStar - Coarse Perception":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMStar - Fine-Grained Perception":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMStar - Instance Reasoning":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMStar - Logical Reasoning":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMStar - Math":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMStar - Science and Technology":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - 3D":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - 3D CAD Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - 3D Indoor Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - AND":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - AR":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - AUD":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Abstract Visual Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Action Quality Assessment":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Age, Gender, and Race Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Anatomy Identification":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Animal Keypoint Detection":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Animals Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Animated Character Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Art Design":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Artwork Emotion Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Astronomical Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Attribute Hallucination":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Behavior Anomaly Detection":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Body Emotion Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Building Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Business":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - CIM":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Camouflage Object Detection":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Celebrity Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Chart To Table":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Chart To Text":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Chart VQA":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Chemical Apparatus Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Clock Reading":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Clothes Keypoint Detection":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Color Assimilation":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Color Constancy":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Color Contrast":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Color Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Count":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Counting By Category":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Counting By Reasoning":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Counting By Visual Prompting":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Crowd Counting":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - DKR":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - DU":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Deepfake Detection":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Depth Estimation":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Disaster Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Disease Diagnosis":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - DocVQA":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Electronic Object Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Emo":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Equation-to-LaTeX":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Existence Hallucination":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Face Detection":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Face Mask Anomaly Detection":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Face Retrieval":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Facial Expression Change Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Facial Expression Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Fashion Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Film and Television Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Font Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Food Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Furniture Keypoint Detection":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - GN":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - GUI General":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - GUI Install":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Gaze Estimation":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - General Action Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Geometrical Perspective":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Geometrical Relativity":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Gesture Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Google Apps":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - HLN":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Handwritten Mathematical Expression Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Handwritten Retrieval":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Handwritten Text Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Health and Medicine":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Helmet Anomaly Detection":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Human Interaction Understanding":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Human Keypoint Detection":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Human Object Interaction Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Humanities and Social Science":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - I2IT":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - IEJ":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - IR":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Image Based Action Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Image Captioning":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Image Captioning Paragraph":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Image Colorization":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Image Dense Captioning":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Image Matting":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Image Quality Assessment":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Image Season Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Image-to-Image Retrieval":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Industrial Produce Anomaly Detection":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Instance Captioning":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Interactive Segmentation":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Jigsaw Puzzle Solving":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - KD":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - LVLM Response Judgement":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Landmark Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Lesion Grading":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Loc":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Logo and Brand Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - MIA":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - MeViS":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - MedU":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Medical Modality Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - MemU":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Meme Image Understanding":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Meme Video Understanding":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Micro Expression Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Multiple Image Captioning":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Multiple Instance Captioning":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Multiple View Image Understanding":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Musical Instrument Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - National Flag Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Navigation":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Next Image Prediction":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - OCR":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Object Detection":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - One Shot Detection":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Order Hallucination":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Other Biological Attributes":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - PLP":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Painting Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Person Re-Identification":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Pixel Localization":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Pixel Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Plant Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Point Tracking":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Polygon Localization":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Profession Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - RR":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Raven Progressive Matrices":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Reasoning Segmentation":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Referring Detection":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Relation Hallucination":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Religious Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Remote Sensing Object Detection":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Rock Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Rotated Object Detection":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Salient Object Detection RGB":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Salient Object Detection RGBD":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Scene Emotion Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Scene Graph Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Scene Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Scene Text Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Science":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Screenshot-to-Code":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Sculpture Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Set-of-Mark Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Shape Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Sign Language Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Single Object Tracking":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Sketch-to-Code":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Sketch-to-Image Retrieval":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Small Object Detection":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Social Relation Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Sports Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Spot The Diff":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Spot The Similarity":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - TU":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Table Structure Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Tech Engineering":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Temporal Anticipation":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Temporal Localization":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Temporal Ordering":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Temporal Sequence Understanding":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Text-to-Image Retrieval":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Texture Material Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Traffic Anomaly Detection":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Traffic Light Understanding":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Traffic Participants Understanding":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Traffic Sign Understanding":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Transparent Object Detection":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - VC":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - VG":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - VI":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - VP":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - VPU":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - VR":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Vehicle Keypoint Detection":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Vehicle Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Vehicle Retrieval":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Video Captioning":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Visual Document Information Extraction":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Visual Prompt Understanding":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Waste Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Weapon Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Weather Recognition":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Web Shopping":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Whoops":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MMT-Bench - Writing Poetry From Image":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MTVQA":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MathVista - ALG":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MathVista - ARI":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MathVista - FQA":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MathVista - GEO":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MathVista - GPS":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MathVista - LOG":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MathVista - MWP":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MathVista - NUM":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MathVista - SCI":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MathVista - STA":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MathVista - TQA":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM MathVista - VQA":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM OCR-VQA":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM OCRBench":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Score (normalized)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM POPE":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Overall (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM SEED-Bench 2":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM SEED-Bench 2 Plus":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM ScienceQA Test":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM Subjective":{"url":"https://huggingface.co/spaces/opencompass/openlmm_subjective_leaderboard","metric":"Average Normalized Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM VCR":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Overall Jaccard (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM Video":{"url":"https://huggingface.co/spaces/opencompass/openvlm_video_leaderboard","metric":"Average Normalized Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM Video - MLVU":{"url":"https://huggingface.co/spaces/opencompass/openvlm_video_leaderboard","metric":"Normalized Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM Video - MMBench-Video":{"url":"https://huggingface.co/spaces/opencompass/openvlm_video_leaderboard","metric":"Normalized Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM Video - MVBench":{"url":"https://huggingface.co/spaces/opencompass/openvlm_video_leaderboard","metric":"Normalized Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM Video - TempCompass":{"url":"https://huggingface.co/spaces/opencompass/openvlm_video_leaderboard","metric":"Normalized Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OpenVLM Video - Video-MME (w/o subs)":{"url":"https://huggingface.co/spaces/opencompass/openvlm_video_leaderboard","metric":"Normalized Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Opper TaskBench":{"url":"https://opper.ai/real-world-benchmarks","metric":"Avg Task Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Opus Magnum Bench":{"url":"https://opusmagnumbench.com/","metric":"Human-normalized score (%)","theoretical_min":0.0,"theoretical_max":100.0},"OrchestrationBench":{"url":"https://github.com/kakao/OrchestrationBench","metric":"Average Score","theoretical_min":0.0,"theoretical_max":100.0},"OrgForge-IT":{"url":"https://benchmarklist.com/benchmarks/orgforge_it/","metric":"Verdict F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Organic Chemistry (Anthropic)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Organic Chemistry (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"P2PCLAW Innovative Benchmark":{"url":"https://huggingface.co/spaces/Agnuxo/P2PCLAW-Benchmark","metric":"Average Score","theoretical_min":0.0,"theoretical_max":10.0},"P3B3":{"url":"https://benchmarklist.com/benchmarks/p3b3/","metric":"LLM pt-PT (↑) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"PACT (Lechmazur)":{"url":"https://github.com/lechmazur/pact","metric":"CMS Points","theoretical_min":0.0,"theoretical_max":null},"PARROT - Dialect Compatibility":{"url":"https://code4db.github.io/parrot-bench/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"PARROT - MySQL Dialect Compatibility":{"url":"https://code4db.github.io/parrot-bench/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"PARROT - Oracle Dialect Compatibility":{"url":"https://code4db.github.io/parrot-bench/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"PARROT - Result Consistency":{"url":"https://code4db.github.io/parrot-bench/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - Basic Wiring Error Identification":{"url":"https://digailab.github.io/PCB-Bench/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - Common Basic CQ":{"url":"https://digailab.github.io/PCB-Bench/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - Common Basic Fill in Blank":{"url":"https://digailab.github.io/PCB-Bench/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - Placement Macro CQ":{"url":"https://digailab.github.io/PCB-Bench/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - Placement Macro QA BERTScore":{"url":"https://digailab.github.io/PCB-Bench/","metric":"BERTScore (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - Placement Macro QA SBERT":{"url":"https://digailab.github.io/PCB-Bench/","metric":"SBERT similarity (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - Placement Micro CQ":{"url":"https://digailab.github.io/PCB-Bench/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - Placement Micro QA BERTScore":{"url":"https://digailab.github.io/PCB-Bench/","metric":"BERTScore (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - Placement Micro QA SBERT":{"url":"https://digailab.github.io/PCB-Bench/","metric":"SBERT similarity (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - QA Placement BERTScore":{"url":"https://digailab.github.io/PCB-Bench/","metric":"BERTScore (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - QA Placement SBERT":{"url":"https://digailab.github.io/PCB-Bench/","metric":"SBERT similarity (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - QA Routing BERTScore":{"url":"https://digailab.github.io/PCB-Bench/","metric":"BERTScore (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - QA Routing SBERT":{"url":"https://digailab.github.io/PCB-Bench/","metric":"SBERT similarity (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - Routing Macro CQ":{"url":"https://digailab.github.io/PCB-Bench/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - Routing Macro QA BERTScore":{"url":"https://digailab.github.io/PCB-Bench/","metric":"BERTScore (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - Routing Macro QA SBERT":{"url":"https://digailab.github.io/PCB-Bench/","metric":"SBERT similarity (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - Routing Micro CQ":{"url":"https://digailab.github.io/PCB-Bench/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - Routing Micro QA BERTScore":{"url":"https://digailab.github.io/PCB-Bench/","metric":"BERTScore (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - Routing Micro QA SBERT":{"url":"https://digailab.github.io/PCB-Bench/","metric":"SBERT similarity (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - Task 3 BERTScore":{"url":"https://digailab.github.io/PCB-Bench/","metric":"BERTScore (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - Task 3 F1":{"url":"https://digailab.github.io/PCB-Bench/","metric":"F1 score (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - Task 3 Precision":{"url":"https://digailab.github.io/PCB-Bench/","metric":"Precision (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - Task 3 Recall":{"url":"https://digailab.github.io/PCB-Bench/","metric":"Recall (%)","theoretical_min":0.0,"theoretical_max":100.0},"PCB-Bench - Task 3 SBERT":{"url":"https://digailab.github.io/PCB-Bench/","metric":"SBERT similarity (%)","theoretical_min":0.0,"theoretical_max":100.0},"PDP-Bench":{"url":"https://benchmarklist.com/benchmarks/pdp_bench/","metric":"Macro-F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"PECC":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"Pass@3","theoretical_min":0.0,"theoretical_max":100.0},"PECC - Advent of Code":{"url":"https://huggingface.co/spaces/PatrickHaller/pecc-leaderboard","metric":"Solve rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"PECC - Advent of Code (LeetCode-style)":{"url":"https://huggingface.co/spaces/PatrickHaller/pecc-leaderboard","metric":"Solve rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"PECC - Project Euler":{"url":"https://huggingface.co/spaces/PatrickHaller/pecc-leaderboard","metric":"Solve rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"PECC - Project Euler (Story)":{"url":"https://huggingface.co/spaces/PatrickHaller/pecc-leaderboard","metric":"Solve rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"PHYBench":{"url":"https://www.phybench.cn/leaderboard","metric":"EED Score","theoretical_min":null,"theoretical_max":null},"PIQA":{"url":"https://epoch.ai/benchmarks/piqa","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":50},"PL-MTEB":{"url":"https://huggingface.co/spaces/PL-MTEB/leaderboard","metric":"Mean Main Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"PLawBench":{"url":"https://benchmarklist.com/benchmarks/plawbench/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"PM-AGI":{"url":"https://huggingface.co/spaces/Hawky-ai/pm-agi-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"PM-AGI - Category - Action Based":{"url":"https://huggingface.co/spaces/Hawky-ai/pm-agi-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"PM-AGI - Category - Critical Thinking":{"url":"https://huggingface.co/spaces/Hawky-ai/pm-agi-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"PM-AGI - Category - Google Ads":{"url":"https://huggingface.co/spaces/Hawky-ai/pm-agi-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"PM-AGI - Category - Meta Ads":{"url":"https://huggingface.co/spaces/Hawky-ai/pm-agi-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"PM-AGI - Difficulty - Easy":{"url":"https://huggingface.co/spaces/Hawky-ai/pm-agi-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"PM-AGI - Difficulty - Hard":{"url":"https://huggingface.co/spaces/Hawky-ai/pm-agi-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"PM-AGI - Difficulty - Medium":{"url":"https://huggingface.co/spaces/Hawky-ai/pm-agi-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"PM-AGI - Question Type - Action Based":{"url":"https://huggingface.co/spaces/Hawky-ai/pm-agi-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"PM-AGI - Question Type - Mcq":{"url":"https://huggingface.co/spaces/Hawky-ai/pm-agi-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"PM-AGI - Reasoning Type - Adversarial":{"url":"https://huggingface.co/spaces/Hawky-ai/pm-agi-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"PM-AGI - Reasoning Type - Creative Strategy":{"url":"https://huggingface.co/spaces/Hawky-ai/pm-agi-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"PM-AGI - Reasoning Type - Diagnostic":{"url":"https://huggingface.co/spaces/Hawky-ai/pm-agi-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"PM-AGI - Reasoning Type - Quantitative":{"url":"https://huggingface.co/spaces/Hawky-ai/pm-agi-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"PM-AGI - Reasoning Type - Recall":{"url":"https://huggingface.co/spaces/Hawky-ai/pm-agi-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"PM-LLM-Benchmark":{"url":"https://github.com/fit-alessandro-berti/pm-llm-benchmark","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"POLAR-Bench":{"url":"https://benchmarklist.com/benchmarks/polar_bench_a_diagnostic_benchmark_for_privacy_utility_trade_offs_in_llm_agents/","metric":"Overall Mean (self-reported)","theoretical_min":0.0,"theoretical_max":null},"PRMBench":{"url":"https://prmbench.github.io/","metric":"Overall Score","theoretical_min":0.0,"theoretical_max":100.0},"PaliBench":{"url":"https://benchmarklist.com/benchmarks/palibench/","metric":"sim_best (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Pangram AI-Text Detection Hard Set (8-shot)":{"url":"https://www.pangram.com/blog/ai-has-an-eye-for-ai","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pangram AI-Text Detection Hard Set (zero-shot)":{"url":"https://www.pangram.com/blog/ai-has-an-eye-for-ai","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"PaperBench":{"url":"https://github.com/openai/preparedness/tree/main/project/paperbench","metric":"Replication Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"PaperFit-Bench":{"url":"https://benchmarklist.com/benchmarks/paperfit_bench/","metric":"Compile (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ParallelKernelBench":{"url":"https://www.together.ai/blog/parallelkernelbench","metric":"Fast1@3 (% of problems)","theoretical_min":0.0,"theoretical_max":100.0},"ParallelKernelBench Fast1@1":{"url":"https://www.together.ai/blog/parallelkernelbench","metric":"Fast1@1 (% of problems)","theoretical_min":0.0,"theoretical_max":100.0},"ParallelKernelBench Pass@1":{"url":"https://www.together.ai/blog/parallelkernelbench","metric":"Pass@1 (% of problems)","theoretical_min":0.0,"theoretical_max":100.0},"ParallelKernelBench Pass@3":{"url":"https://www.together.ai/blog/parallelkernelbench","metric":"Pass@3 (% of problems)","theoretical_min":0.0,"theoretical_max":100.0},"Paraphrase-Induced Output-Mode Collapse":{"url":"https://benchmarklist.com/benchmarks/paraphrase_induced_output_mode_collapse_when_llms_break_character_under_semantically_equivalent_inputs/","metric":"Semantic Consistency Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ParseBench":{"url":"https://parsebench.ai/","metric":"Overall Score","theoretical_min":null,"theoretical_max":null},"PassBench":{"url":"https://benchmarklist.com/benchmarks/passbench/","metric":"AS Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"PawBench - Hermes":{"url":"https://agentscope-ai.github.io/PawBench/","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"PawBench - OpenClaw":{"url":"https://agentscope-ai.github.io/PawBench/","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"PawBench - QwenPaw":{"url":"https://agentscope-ai.github.io/PawBench/","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pencil Puzzle Bench - Country":{"url":"https://ppbench.com/","metric":"Direct-ask Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pencil Puzzle Bench - Double Choco":{"url":"https://ppbench.com/","metric":"Direct-ask Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pencil Puzzle Bench - Firefly":{"url":"https://ppbench.com/","metric":"Direct-ask Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pencil Puzzle Bench - Heyawake":{"url":"https://ppbench.com/","metric":"Direct-ask Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pencil Puzzle Bench - Hitori":{"url":"https://ppbench.com/","metric":"Direct-ask Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pencil Puzzle Bench - Kurodoko":{"url":"https://ppbench.com/","metric":"Direct-ask Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pencil Puzzle Bench - LITS":{"url":"https://ppbench.com/","metric":"Direct-ask Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pencil Puzzle Bench - Light Up":{"url":"https://ppbench.com/","metric":"Direct-ask Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pencil Puzzle Bench - Mashu":{"url":"https://ppbench.com/","metric":"Direct-ask Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pencil Puzzle Bench - Norinori":{"url":"https://ppbench.com/","metric":"Direct-ask Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pencil Puzzle Bench - Nurikabe":{"url":"https://ppbench.com/","metric":"Direct-ask Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pencil Puzzle Bench - Nurimaze":{"url":"https://ppbench.com/","metric":"Direct-ask Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pencil Puzzle Bench - Nurimisaki":{"url":"https://ppbench.com/","metric":"Direct-ask Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pencil Puzzle Bench - Sashigane":{"url":"https://ppbench.com/","metric":"Direct-ask Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pencil Puzzle Bench - Shakashaka":{"url":"https://ppbench.com/","metric":"Direct-ask Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pencil Puzzle Bench - Shikaku":{"url":"https://ppbench.com/","metric":"Direct-ask Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pencil Puzzle Bench - Slitherlink":{"url":"https://ppbench.com/","metric":"Direct-ask Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pencil Puzzle Bench - Sudoku":{"url":"https://ppbench.com/","metric":"Direct-ask Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pencil Puzzle Bench - Tapa":{"url":"https://ppbench.com/","metric":"Direct-ask Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pencil Puzzle Bench - Yajilin":{"url":"https://ppbench.com/","metric":"Direct-ask Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Perception or Prejudice":{"url":"https://benchmarklist.com/benchmarks/perception_or_prejudice_can_mllms_go_beyond_first_impressions_of_personality/","metric":"HR (self-reported)","theoretical_min":0.0,"theoretical_max":null},"PerfCodeBench":{"url":"https://benchmarklist.com/benchmarks/perfcodebench_benchmarking_llms_for_system_level_high_performance_code_optimization/","metric":"CGRE (self-reported)","theoretical_min":0.0,"theoretical_max":null},"PersLitEval":{"url":"https://benchmarklist.com/benchmarks/persliteval_fine_grained_benchmark_and_evaluation_of_llms_on_persian_literature_questions/","metric":"Overall Accuracy (%) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"PersonaArena":{"url":"https://benchmarklist.com/benchmarks/personaarena_dynamic_simulation_for_evaluating_and_enhancing_persona_level_role_playing_in_large_language_models/","metric":"Average (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Personalized Turn-Level User Conversation Sati":{"url":"https://benchmarklist.com/benchmarks/personalized_turn_level_user_conversation_satisfaction_benchmark/","metric":"Micro (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Persuasion (Lechmazur)":{"url":"https://github.com/lechmazur/persuasion","metric":"Average Persuasion Strength","theoretical_min":-6.0,"theoretical_max":6.0},"PhiBench":{"url":"https://benchmarklist.com/benchmarks/phibench/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Phylogenetics (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"PhysBench":{"url":"https://physbench.github.io/","metric":"Overall Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Physical AI Bench - Understanding Overall":{"url":"https://huggingface.co/spaces/shi-labs/physical-ai-bench-leaderboard","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Physical AI Bench Understanding":{"url":"https://huggingface.co/spaces/shi-labs/physical-ai-bench-leaderboard","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Physical Reasoning - CausalVQA":{"url":"https://huggingface.co/spaces/facebook/physical_reasoning_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Physical Reasoning - IntPhys 2":{"url":"https://huggingface.co/spaces/facebook/physical_reasoning_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Physical Reasoning - MVPBench":{"url":"https://huggingface.co/spaces/facebook/physical_reasoning_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Physical Reasoning from Video":{"url":"https://huggingface.co/spaces/facebook/physical_reasoning_leaderboard","metric":"Average Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"PhysicianBench":{"url":"https://benchmarklist.com/benchmarks/physicianbench/","metric":"Pass@1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"PhysicsFinals":{"url":"https://benchmarklist.com/benchmarks/physicsfinals/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"PickMeBench":{"url":"https://pick-me-bench-site.vercel.app/","metric":"Win Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"PinTrace":{"url":"https://benchmarklist.com/benchmarks/pintrace/","metric":"ρ_U(%) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"PinchBench":{"url":"https://pinchbench.com/","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pinocchio Italian - Cultura":{"url":"https://huggingface.co/spaces/mii-llm/pinocchio-ita-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pinocchio Italian - Diritto":{"url":"https://huggingface.co/spaces/mii-llm/pinocchio-ita-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pinocchio Italian - Generale":{"url":"https://huggingface.co/spaces/mii-llm/pinocchio-ita-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pinocchio Italian - Lingua Straniera":{"url":"https://huggingface.co/spaces/mii-llm/pinocchio-ita-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pinocchio Italian - Logica":{"url":"https://huggingface.co/spaces/mii-llm/pinocchio-ita-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pinocchio Italian - Matematica E Scienze":{"url":"https://huggingface.co/spaces/mii-llm/pinocchio-ita-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pinocchio Italian Leaderboard":{"url":"https://huggingface.co/spaces/mii-llm/pinocchio-ita-leaderboard","metric":"Average Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"PlaceboBench":{"url":"https://benchmarklist.com/benchmarks/placebobench/","metric":"Non-Hallucination Rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"PlanBench-XL":{"url":"https://benchmarklist.com/benchmarks/planbench_xl/","metric":"Accuracy (%) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"PlanningBench":{"url":"https://benchmarklist.com/benchmarks/planningbench/","metric":"All-pass (%) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Plant, Persist, Trigger":{"url":"https://benchmarklist.com/benchmarks/plant_persist_trigger_sleeper_attack_on_large_language_model_agents/","metric":"Avg. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"PlantMarkerBench":{"url":"https://benchmarklist.com/benchmarks/plantmarkerbench/","metric":"Valid F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"PlatinumBench (MIT)":{"url":"https://platinum-bench.csail.mit.edu/","metric":"Avg Error Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"PokeAgent Challenge":{"url":"https://pokeagentchallenge.com/","metric":"Elo Rating","theoretical_min":null,"theoretical_max":null},"Poker Agent":{"url":"https://benchmarklist.com/benchmarks/vals_poker_agent/","metric":"TrueSkill (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Poker Arena":{"url":"https://arxiv.org/html/2606.13815v1","metric":"Tournament win rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Polaris-Bench":{"url":"https://benchmarklist.com/benchmarks/polaris_bench/","metric":"Overall C (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Polish EQ-Bench":{"url":"https://huggingface.co/spaces/speakleash/polish_eq-bench","metric":"EQ-Bench Score","theoretical_min":null,"theoretical_max":null},"Position Bias (Lechmazur)":{"url":"https://github.com/lechmazur/position_bias","metric":"Order Flip % (lower is better)","theoretical_min":0.0,"theoretical_max":100.0},"PostTrainBench":{"url":"https://posttrainbench.com/","metric":"Weighted Avg Score","theoretical_min":0.0,"theoretical_max":100.0},"PreScam":{"url":"https://benchmarklist.com/benchmarks/prescam_a_benchmark_for_predicting_scam_progression_from_early_conversations/","metric":"AHit (self-reported)","theoretical_min":0.0,"theoretical_max":null},"PredictionArena (Kalshi)":{"url":"https://www.predictionarena.ai/?platform=kalshi","metric":"Account Value ($)","theoretical_min":0.0,"theoretical_max":null},"PredictionArena (Polymarket)":{"url":"https://www.predictionarena.ai/","metric":"Account Value ($)","theoretical_min":0.0,"theoretical_max":null},"PrepBench":{"url":"https://benchmarklist.com/benchmarks/prepbench/","metric":"End-to-End Prep-Code Acc. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"PrinzBench":{"url":"https://github.com/prinz-ai/prinzbench","metric":"Score (x/99)","theoretical_min":0.0,"theoretical_max":99.0},"PrivacyPeek":{"url":"https://benchmarklist.com/benchmarks/privacypeek/","metric":"TCR (self-reported)","theoretical_min":0.0,"theoretical_max":null},"PrivacySIM":{"url":"https://benchmarklist.com/benchmarks/privacysim/","metric":"acc(E_u+A_u) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ProLLM - Entity Extraction":{"url":"https://www.prollm.ai/leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"ProLLM - Function Calling":{"url":"https://www.prollm.ai/leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"ProLLM - Image Understanding":{"url":"https://www.prollm.ai/leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"ProLLM - LLM-as-a-Judge":{"url":"https://www.prollm.ai/leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"ProLLM - OpenBook Q&A":{"url":"https://www.prollm.ai/leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"ProLLM - Q&A Assistant":{"url":"https://www.prollm.ai/leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"ProLLM - SQL Disambiguation":{"url":"https://www.prollm.ai/leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"ProLLM - StackEval":{"url":"https://www.prollm.ai/leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"ProLLM - StackUnseen":{"url":"https://www.prollm.ai/leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"ProLLM - Summarization":{"url":"https://www.prollm.ai/leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"ProLLM - Transcription":{"url":"https://www.prollm.ai/leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"ProactBench":{"url":"https://benchmarklist.com/benchmarks/proactbench/","metric":"Overall Pass Rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ProfBench":{"url":"https://huggingface.co/spaces/nvidia/ProfBench","metric":"Overall Rubric Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"ProgramBench":{"url":"https://programbench.com/","metric":"Resolved (%)","theoretical_min":0.0,"theoretical_max":100.0},"ProgramBench Almost":{"url":"https://programbench.com/","metric":"Almost (%)","theoretical_min":0.0,"theoretical_max":100.0},"ProphetArena":{"url":"https://www.prophetarena.co/leaderboard","metric":"1 - Brier Score","theoretical_min":0.0,"theoretical_max":1.0},"Prosa":{"url":"https://benchmarklist.com/benchmarks/prosa_rubric_based_evaluation_of_llms_on_real_user_chats_in_brazilian_portuguese/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ProtStructQA":{"url":"https://benchmarklist.com/benchmarks/protstructqa/","metric":"Standard (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ProteinGym Hard":{"url":"https://benchmarklist.com/benchmarks/proteingym_hard/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ProteinGym Hard (Anthropic)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Rank Correlation (%)","theoretical_min":0.0,"theoretical_max":100.0},"Protocol Troubleshooting (Anthropic)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Pseudo-Deliberation in Language Models":{"url":"https://benchmarklist.com/benchmarks/pseudo_deliberation_in_language_models_when_reasoning_fails_to_align_values_and_actions/","metric":"Fast Thinking Macro F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"PubMedQA":{"url":"https://pubmedqa.github.io/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"PutnamBench":{"url":"https://github.com/trishullab/PutnamBench","metric":"Problems Solved","theoretical_min":0.0,"theoretical_max":null},"QUIET":{"url":"https://benchmarklist.com/benchmarks/quiet/","metric":"QUIET Total (self-reported)","theoretical_min":0.0,"theoretical_max":null},"QZhou Flowchart QA":{"url":"https://huggingface.co/datasets/Kingsoft-LLM/QZhou-Flowchart-QA-Benchmark","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qiskit QuantumKatas":{"url":"https://benchmarklist.com/benchmarks/qiskit_quantumkatas_adapting_microsoft_s_quantum_computing_exercises_for_llm_evaluation/","metric":"Best (self-reported)","theoretical_min":0.0,"theoretical_max":null},"QuestBench":{"url":"https://benchmarklist.com/benchmarks/questbench/","metric":"Pass Rate (%) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Qwen-AgentWorld":{"url":"https://benchmarklist.com/benchmarks/qwen_agentworld_language_world_models_for_general_agents/","metric":"Avg. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Qwen3.7 Launch - Apex":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - BFCL-V4":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - ClawEval":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - CoWorkBench":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - CritPT":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - GPQA Diamond":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - Global PIQA":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - HMMT 2026 Feb":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - Humanity's Last Exam":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - Humanity's Last Exam (with tools)":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - IFBench":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - IFEval":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - IMOAnswerBench":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - INCLUDE":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - Kernel Bench L3 - Median Speedup":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Median speedup (x)","theoretical_min":0.0,"theoretical_max":null},"Qwen3.7 Launch - Kernel Bench L3 - Win Rate":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Problems faster than torch.compile (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - LiveCodeBench":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - MAXIFE":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - MCP-Atlas":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - MCP-Mark":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - MMLU-Pro":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - MMLU-ProX":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - MMLU-Redux":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - MMMLU":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - MRCR-v2 128k":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - NL2repo":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - NOVA-63":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - PolyMATH":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - QwenSVG":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Elo","theoretical_min":null,"theoretical_max":null},"Qwen3.7 Launch - QwenWebDev":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Elo","theoretical_min":null,"theoretical_max":null},"Qwen3.7 Launch - QwenWorldBench":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - Qwenclaw":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - SWE-Multilingual":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Resolved (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - SWE-Pro":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Resolved (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - SWE-Verified":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Resolved (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - SciCode":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - Skillsbench":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - SpreadSheetBench-v1":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - SuperGPQA":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - Terminal Bench 2.0-Terminus":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - Vitabench":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Qwen3.7 Launch - WMT24++":{"url":"https://qwen.ai/blog?id=qwen3.7","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"RABBITS":{"url":"https://huggingface.co/spaces/AIM-Harvard/rabbits-leaderboard","metric":"B4BQA Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"RABBITS B4B":{"url":"https://huggingface.co/spaces/AIM-Harvard/rabbits-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"RABBITS B4BQA":{"url":"https://huggingface.co/spaces/AIM-Harvard/rabbits-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"RABBITS MedMCQA G2B":{"url":"https://huggingface.co/spaces/AIM-Harvard/rabbits-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"RABBITS MedMCQA Original":{"url":"https://huggingface.co/spaces/AIM-Harvard/rabbits-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"RABBITS MedQA G2B":{"url":"https://huggingface.co/spaces/AIM-Harvard/rabbits-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"RABBITS MedQA Original":{"url":"https://huggingface.co/spaces/AIM-Harvard/rabbits-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"RAGTruth":{"url":"https://benchmarklist.com/benchmarks/ragtruth/","metric":"RAG Truth (self-reported)","theoretical_min":0.0,"theoretical_max":null},"RBench-V":{"url":"https://evalmodels.github.io/rbenchv/","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"REAL Evals":{"url":"https://www.realevals.xyz/","metric":"Task Completion (%)","theoretical_min":0.0,"theoretical_max":100.0},"REBUS":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"Accuracy","theoretical_min":0.0,"theoretical_max":100.0},"ROK-FORTRESS":{"url":"https://benchmarklist.com/benchmarks/rok_fortress/","metric":"Δ_ling (pp) (self-reported)","theoretical_min":null,"theoretical_max":null},"ROSE":{"url":"https://benchmarklist.com/benchmarks/rose_benchmarking_the_perception_to_action_gap_in_multimodal_models/","metric":"Avg. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"RP-Bench":{"url":"https://huggingface.co/datasets/lazyweasel/roleplay-bench","metric":"Combined Score","theoretical_min":0.0,"theoretical_max":100.0},"RP-Bench - Elo":{"url":"https://huggingface.co/datasets/lazyweasel/roleplay-bench","metric":"Elo","theoretical_min":null,"theoretical_max":null},"RP-Bench - Flaw Hunter":{"url":"https://huggingface.co/datasets/lazyweasel/roleplay-bench","metric":"Flaw Hunter Score","theoretical_min":0.0,"theoretical_max":100.0},"RP-Bench - Objective":{"url":"https://huggingface.co/datasets/lazyweasel/roleplay-bench","metric":"Objective Score","theoretical_min":0.0,"theoretical_max":100.0},"RP-Leaderboard":{"url":"https://psychosmiley.github.io/RP-Leaderboard/","metric":"RP Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"RPC-Bench":{"url":"https://rpc-bench.github.io/","metric":"F1-like (%)","theoretical_min":0.0,"theoretical_max":100.0},"RULER":{"url":"https://github.com/NVIDIA/RULER","metric":"Avg Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"RadLE 2.0 (Radiology)":{"url":"https://crashlab.in/radle/2/leaderboard","metric":"RadLE Score","theoretical_min":null,"theoretical_max":null},"Ramp SWE-Bench":{"url":"https://labs.ramp.com/swebench","metric":"Resolved (%)","theoretical_min":0.0,"theoretical_max":100.0},"RankJudge":{"url":"https://benchmarklist.com/benchmarks/rankjudge_a_multi_turn_llm_as_a_judge_synthetic_benchmark_generator/","metric":"Elo (All) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"React Native Evals":{"url":"https://rn-evals.vercel.app/","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Real-World Finance v1":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Real-World Finance v2":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Elo","theoretical_min":null,"theoretical_max":null},"RealDataAgentBench":{"url":"https://patibandlavenkatamanideep.github.io/RealDataAgentBench/","metric":"Average DAB Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"RealWorldQA":{"url":"https://benchmarklist.com/benchmarks/realworldqa/","metric":"RealWorldQA (self-reported)","theoretical_min":0.0,"theoretical_max":null},"RealityTest":{"url":"https://benchmarklist.com/benchmarks/realitytest/","metric":"Text disclosure probability (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ReasonScape M12X":{"url":"https://reasonscape.com/m12x/leaderboard/","metric":"ReasonScore (hard)","theoretical_min":null,"theoretical_max":null},"ReasonScape R12":{"url":"https://reasonscape.com/r12/leaderboard/","metric":"ReasonScore","theoretical_min":null,"theoretical_max":null},"RecoAtlas":{"url":"https://benchmarklist.com/benchmarks/recoatlas_from_semantic_plausibility_to_set_level_utility_in_llm_recommendation_agents/","metric":"Bundle SetHit@20 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"RedCode":{"url":"https://huggingface.co/spaces/FrontierAICybersecurity/Cybersecurity_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"RedSage-Bench":{"url":"https://benchmarklist.com/benchmarks/redsage_bench/","metric":"Macro Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"RefSpatialBench":{"url":"https://benchmarklist.com/benchmarks/refspatialbench/","metric":"RefSpatialBench (self-reported)","theoretical_min":0.0,"theoretical_max":null},"RefusalBench":{"url":"https://benchmarklist.com/benchmarks/refusalbench/","metric":"Strict Refusal Rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Relevance as a Vulnerability":{"url":"https://benchmarklist.com/benchmarks/relevance_as_a_vulnerability_how_web_retrieval_degrades_safety_alignment_in_llm_agents/","metric":"Agent (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ReliableMath - Precision":{"url":"https://huggingface.co/spaces/BeyondHsueh/ReliableMath-Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"ReliableMath - Prudence":{"url":"https://huggingface.co/spaces/BeyondHsueh/ReliableMath-Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"RepairBench":{"url":"https://repairbench.github.io/","metric":"Plausible@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"RepoMirage":{"url":"https://benchmarklist.com/benchmarks/repomirage_probing_repository_context_reasoning_in_code_agents_with_perturbations/","metric":"Avg. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"RepoQA":{"url":"https://benchmarklist.com/benchmarks/repoqa/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ReproRepo":{"url":"https://benchmarklist.com/benchmarks/reprorepo/","metric":"Issue Match EM@10 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ResearchClawBench":{"url":"https://benchmarklist.com/benchmarks/researchclawbench_a_benchmark_for_end_to_end_autonomous_scientific_research/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ResearchCodeBench":{"url":"https://researchcodebench.github.io/leaderboard/","metric":"Task Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"RevengeBench":{"url":"https://revengebench.com/","metric":"Arena Mean Score","theoretical_min":null,"theoretical_max":null},"Reward Hacking Benchmark (RHB)":{"url":"https://benchmarklist.com/benchmarks/reward_hacking_benchmark_rhb/","metric":"Exploit rate (%) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"RewardBench":{"url":"https://huggingface.co/datasets/allenai/reward-bench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"RewardBench 2 Factuality":{"url":"https://huggingface.co/datasets/allenai/reward-bench-2","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"RewardBench 2 Focus":{"url":"https://huggingface.co/datasets/allenai/reward-bench-2","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"RewardBench 2 Math":{"url":"https://huggingface.co/datasets/allenai/reward-bench-2","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"RewardBench 2 Precise IF":{"url":"https://huggingface.co/datasets/allenai/reward-bench-2","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"RewardBench 2 Safety":{"url":"https://huggingface.co/datasets/allenai/reward-bench-2","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"RewardBench Chat":{"url":"https://huggingface.co/datasets/allenai/reward-bench","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"RewardBench Chat Hard":{"url":"https://huggingface.co/datasets/allenai/reward-bench","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"RewardBench Factuality":{"url":"https://huggingface.co/datasets/allenai/reward-bench-2-results","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"RewardBench Focus":{"url":"https://huggingface.co/datasets/allenai/reward-bench-2-results","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"RewardBench Math":{"url":"https://huggingface.co/datasets/allenai/reward-bench-2-results","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"RewardBench Precise IF":{"url":"https://huggingface.co/datasets/allenai/reward-bench-2-results","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"RewardBench Prior Sets (0.5 weight)":{"url":"https://huggingface.co/datasets/allenai/reward-bench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"RewardBench Reasoning":{"url":"https://huggingface.co/datasets/allenai/reward-bench","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"RewardBench Safety":{"url":"https://huggingface.co/datasets/allenai/reward-bench","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"RewardBench Ties":{"url":"https://huggingface.co/datasets/allenai/reward-bench-2-results","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Riemann-bench":{"url":"https://surgehq.ai/leaderboards","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"RiemannBench (Fable/Mythos)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"RoadmapBench":{"url":"https://benchmarklist.com/benchmarks/roadmapbench/","metric":"Resolved (%) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"RoboBench":{"url":"https://benchmarklist.com/benchmarks/robobench/","metric":"Overall Dimension Average (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Robusto-2":{"url":"https://benchmarklist.com/benchmarks/robusto_2_benchmarking_humans_vlms_for_autonomous_driving_in_lima_new_york_city/","metric":"Aggregate Accuracy (Reasoning) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Rogo Big Finance Bench":{"url":"https://benchmarklist.com/benchmarks/rogo_big_finance_benchmark/","metric":"Rubric Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Ru Arena Hard":{"url":"https://huggingface.co/spaces/vikhrmodels/arenahardlb","metric":"Win Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"RubberDuckBench":{"url":"https://cs.brynmawr.edu/RubberDuckBench/","metric":"Performance (%)","theoretical_min":0.0,"theoretical_max":100.0},"RubricEval":{"url":"https://huggingface.co/spaces/vbhat4/rubriceval","metric":"Mean Score (1-5)","theoretical_min":1.0,"theoretical_max":5.0},"RuneBench":{"url":"https://maxbittker.github.io/runebench/","metric":"Total Peak XP Rate (XP/min)","theoretical_min":0.0,"theoretical_max":null},"RxEval":{"url":"https://benchmarklist.com/benchmarks/rxeval_a_prescription_level_benchmark_for_evaluating_llm_medication_recommendation/","metric":"F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"SAFIM":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"Average","theoretical_min":0.0,"theoretical_max":100.0},"SAGE":{"url":"https://benchmarklist.com/benchmarks/sage_scalable_automated_robustness_augmentation_for_llm_knowledge_evaluation/","metric":"OA (%) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"SALAD-Bench":{"url":"https://huggingface.co/spaces/opensafetylab/Salad-Bench-Leaderboard","metric":"Average Safety Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"SALAD-Bench Attack":{"url":"https://huggingface.co/spaces/opensafetylab/Salad-Bench-Leaderboard","metric":"Safety Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"SALAD-Bench Base":{"url":"https://huggingface.co/spaces/opensafetylab/Salad-Bench-Leaderboard","metric":"Safety Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"SCICONVBENCH":{"url":"https://benchmarklist.com/benchmarks/sciconvbench_benchmarking_llms_on_multi_turn_clarification_for_task_formulation_in_computational_science/","metric":"FRR (self-reported)","theoretical_min":0.0,"theoretical_max":null},"SCORE Robustness (Accuracy)":{"url":"https://huggingface.co/spaces/nvidia/llm-robustness-leaderboard","metric":"Average Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SCORE Robustness (Consistency)":{"url":"https://huggingface.co/spaces/nvidia/llm-robustness-leaderboard","metric":"Average Consistency Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"SEA LLM Leaderboard - SeaBench":{"url":"https://huggingface.co/spaces/SeaLLMs/LLM_Leaderboard_for_SEA","metric":"Private Average Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"SEA LLM Leaderboard - SeaExam":{"url":"https://huggingface.co/spaces/SeaLLMs/LLM_Leaderboard_for_SEA","metric":"Private Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"SEA-HELM":{"url":"https://leaderboard.sea-lion.ai/","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - Adversarial Robustness":{"url":"https://scale.com/leaderboard/adversarial_robustness","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - Agentic Tool Use (Chat)":{"url":"https://scale.com/leaderboard/tool_use_chat","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - Agentic Tool Use (Enterprise)":{"url":"https://scale.com/leaderboard/tool_use_enterprise","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - Arabic":{"url":"https://scale.com/leaderboard/arabic","metric":"Score","theoretical_min":null,"theoretical_max":null},"SEAL - AudioMultiChallenge":{"url":"https://scale.com/leaderboard/audiomc","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - AudioMultiChallenge - Text Output":{"url":"https://scale.com/leaderboard/audiomc-text","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - Chinese":{"url":"https://scale.com/leaderboard/chinese","metric":"Score","theoretical_min":null,"theoretical_max":null},"SEAL - Coding":{"url":"https://scale.com/leaderboard/coding","metric":"Score","theoretical_min":null,"theoretical_max":null},"SEAL - EnigmaEval":{"url":"https://scale.com/leaderboard/enigma_eval","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0,"sources":[{"name":"SEAL - EnigmaEval","url":"https://scale.com/leaderboard/enigma_eval","metric":"Score"},{"name":"Epoch AI - Enigma Eval","url":"https://epoch.ai/benchmarks/enigma-eval","metric":"Score"}]},"SEAL - Fortress":{"url":"https://scale.com/leaderboard/fortress","metric":"Score","theoretical_min":null,"theoretical_max":null},"SEAL - Humanity's Last Exam":{"url":"https://scale.com/leaderboard/humanitys_last_exam","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - Instruction Following":{"url":"https://scale.com/leaderboard/instruction_following","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - Japanese":{"url":"https://scale.com/leaderboard/japanese","metric":"Score","theoretical_min":null,"theoretical_max":null},"SEAL - Korean":{"url":"https://scale.com/leaderboard/korean","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - MASK":{"url":"https://scale.com/leaderboard/mask","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - MCP Atlas":{"url":"https://scale.com/leaderboard/mcp_atlas","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - Math":{"url":"https://scale.com/leaderboard/math","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - MultiChallenge":{"url":"https://scale.com/leaderboard/multichallenge","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - MultiNRC":{"url":"https://scale.com/leaderboard/multinrc","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - Professional Reasoning Benchmark - Finance":{"url":"https://scale.com/leaderboard/prbench-finance","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - Professional Reasoning Benchmark - Legal":{"url":"https://scale.com/leaderboard/prbench-legal","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - PropensityBench":{"url":"https://scale.com/leaderboard/propensitybench","metric":"Score","theoretical_min":null,"theoretical_max":null},"SEAL - Remote Labor Index (RLI)":{"url":"https://scale.com/leaderboard/rli","metric":"Score","theoretical_min":null,"theoretical_max":null},"SEAL - SWE Atlas - Codebase QnA":{"url":"https://scale.com/leaderboard/sweatlas-qna","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - SWE Atlas - Refactoring":{"url":"https://scale.com/leaderboard/sweatlas-refactoring","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - SWE Atlas - Test Writing":{"url":"https://scale.com/leaderboard/sweatlas-tw","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - SWE-Bench Pro (Private Dataset)":{"url":"https://scale.com/leaderboard/swe_bench_pro_private","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - SWE-Bench Pro (Public Dataset)":{"url":"https://scale.com/leaderboard/swe_bench_pro_public","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - SciPredict":{"url":"https://scale.com/leaderboard/scipredict","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - Spanish":{"url":"https://scale.com/leaderboard/spanish","metric":"Score","theoretical_min":null,"theoretical_max":null},"SEAL - TutorBench":{"url":"https://scale.com/leaderboard/tutorbench","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - VISTA":{"url":"https://scale.com/leaderboard/visual_language_understanding","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL - VisualToolBench (VTB)":{"url":"https://scale.com/leaderboard/vtb","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"SEAL Showdown":{"url":"https://scale.com/showdown","metric":"Arena Score","theoretical_min":null,"theoretical_max":null},"SECQUE":{"url":"https://benchmarklist.com/benchmarks/secque/","metric":"Strict Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"SEEDBench IMG":{"url":"https://huggingface.co/spaces/opencompass/open_vlm_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SGI-Bench":{"url":"https://github.com/InternScience/SGI-Bench","metric":"SGI-Score","theoretical_min":0.0,"theoretical_max":100.0},"SGI-Bench Deep Research":{"url":"https://github.com/InternScience/SGI-Bench","metric":"Deep Research Score","theoretical_min":0.0,"theoretical_max":100.0},"SGI-Bench Dry Experiment":{"url":"https://github.com/InternScience/SGI-Bench","metric":"Dry Experiment Score","theoretical_min":0.0,"theoretical_max":100.0},"SGI-Bench Experimental Reasoning":{"url":"https://github.com/InternScience/SGI-Bench","metric":"Experimental Reasoning Score","theoretical_min":0.0,"theoretical_max":100.0},"SGI-Bench Idea Generation":{"url":"https://github.com/InternScience/SGI-Bench","metric":"Idea Generation Score","theoretical_min":0.0,"theoretical_max":100.0},"SGI-Bench Wet Experiment":{"url":"https://github.com/InternScience/SGI-Bench","metric":"Wet Experiment Score","theoretical_min":0.0,"theoretical_max":100.0},"SGR-Bench":{"url":"https://benchmarklist.com/benchmarks/sgr_bench_benchmarking_search_agents_on_state_gated_retrieval/","metric":"Item-F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"SIBench":{"url":"https://sibench.github.io/Awesome-Visual-Spatial-Reasoning/","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"SLM-RAG Arena":{"url":"https://huggingface.co/spaces/aizip-dev/SLM-RAG-Arena","metric":"Elo Rating","theoretical_min":null,"theoretical_max":null},"SLMJury":{"url":"https://anishh15.github.io/SLMJury/","metric":"Judge accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SLR-Bench - Basic":{"url":"https://huggingface.co/spaces/AIML-TUDA/slr-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SLR-Bench - Easy":{"url":"https://huggingface.co/spaces/AIML-TUDA/slr-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SLR-Bench - Hard":{"url":"https://huggingface.co/spaces/AIML-TUDA/slr-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SLR-Bench - Medium":{"url":"https://huggingface.co/spaces/AIML-TUDA/slr-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SMDD-Bench":{"url":"https://smddbench.com/","metric":"Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"SMDD-Bench Diversity":{"url":"https://smddbench.com/","metric":"Avg Successful","theoretical_min":0.0,"theoretical_max":10.0},"SQL Capability - Dialect Conversion":{"url":"https://sql-llm-leaderboard.com/ranking/2026-06","metric":"Ability Score","theoretical_min":0.0,"theoretical_max":100.0},"SQL Capability - SQL Optimization":{"url":"https://sql-llm-leaderboard.com/ranking/2026-06","metric":"Ability Score","theoretical_min":0.0,"theoretical_max":100.0},"SQL Capability - SQL Understanding":{"url":"https://sql-llm-leaderboard.com/ranking/2026-06","metric":"Ability Score","theoretical_min":0.0,"theoretical_max":100.0},"SQL Capability Leaderboard":{"url":"https://sql-llm-leaderboard.com/ranking/2026-06","metric":"Average Ability Score","theoretical_min":0.0,"theoretical_max":100.0},"SQuAD v1.1":{"url":"https://rajpurkar.github.io/SQuAD-explorer/","metric":"F1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"SQuAD v2.0":{"url":"https://rajpurkar.github.io/SQuAD-explorer/","metric":"F1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"SRE Skills Bench":{"url":"https://sreskillsbench.com/#leaderboard","metric":"Average (%)","theoretical_min":0.0,"theoretical_max":100.0},"SRE Skills Bench - GMCQ":{"url":"https://sreskillsbench.com/#leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SRE Skills Bench - IAM":{"url":"https://sreskillsbench.com/#leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SRE Skills Bench - Kubernetes":{"url":"https://sreskillsbench.com/#leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SRE Skills Bench - S3":{"url":"https://sreskillsbench.com/#leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SRE Skills Bench - Storage":{"url":"https://sreskillsbench.com/#leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SRE Skills Bench - VPC":{"url":"https://sreskillsbench.com/#leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"STALE":{"url":"https://benchmarklist.com/benchmarks/stale/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"STT-Arena":{"url":"https://benchmarklist.com/benchmarks/stt_arena/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"SVFSearch":{"url":"https://benchmarklist.com/benchmarks/svfsearch_a_multimodal_knowledge_intensive_benchmark_for_short_video_frame_search_in_the_gaming_vertical_domain/","metric":"Overall Acc. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"SVLAT - Animation":{"url":"https://github.com/patdmp/mllm-scivis-lit-benchmark","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SVLAT - Image":{"url":"https://github.com/patdmp/mllm-scivis-lit-benchmark","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SVLAT - Overall":{"url":"https://github.com/patdmp/mllm-scivis-lit-benchmark","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SWE Atlas":{"url":"https://benchmarklist.com/benchmarks/sweatlas_codebase_qna/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"SWE-Arena":{"url":"https://swe-arena.com/","metric":"Elo Score","theoretical_min":null,"theoretical_max":null,"sources":[{"name":"SWE-Arena","url":"https://swe-arena.com/","metric":"Elo Score"},{"name":"Software Engineering Arena","url":"https://huggingface.co/spaces/SE-Arena/Software-Engineering-Arena","metric":"Elo Rating"}]},"SWE-Lancer":{"url":"https://www.vals.ai/benchmarks","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"SWE-Lancer (Diamond)":{"url":"https://swelancer.github.io/leaderboard/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"SWE-Marathon":{"url":"https://www.swe-marathon.org/","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"SWE-Milestone":{"url":"https://swe-milestone.com/","metric":"Milestone Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"SWE-PRBench":{"url":"https://benchmarklist.com/benchmarks/swe_prbench/","metric":"Overall (sbar) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"SWE-Together":{"url":"https://togetherbench.com/","metric":"Jury Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"SWE-bench Multilingual (Anthropic Scaffold)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Resolved (%)","theoretical_min":0.0,"theoretical_max":100.0},"SWE-bench Multilingual (Mythos Preview System Card)":{"url":"https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf","metric":"Resolved (%)","theoretical_min":0.0,"theoretical_max":100.0},"SWE-bench Multimodal (Anthropic Internal Harness)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Resolved (%)","theoretical_min":0.0,"theoretical_max":100.0},"SWE-bench Multimodal (Mythos Preview System Card)":{"url":"https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf","metric":"Resolved (%)","theoretical_min":0.0,"theoretical_max":100.0},"SWE-bench Pro (Anthropic Scaffold)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"SWE-bench Verified":{"url":"https://www.swebench.com/","metric":"Resolved (%)","theoretical_min":0.0,"theoretical_max":100.0},"SWE-bench Verified (Anthropic Scaffold)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Resolved (%)","theoretical_min":0.0,"theoretical_max":100.0},"SWE-bench Verified (Fable/Mythos)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Resolved (%)","theoretical_min":0.0,"theoretical_max":100.0},"SWE-bench Verified (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Resolved (%)","theoretical_min":0.0,"theoretical_max":100.0},"SWE-rebench":{"url":"https://swe-rebench.com","metric":"Resolved (%)","theoretical_min":0.0,"theoretical_max":100.0},"SaaS-Bench":{"url":"https://benchmarklist.com/benchmarks/saas_bench_can_computer_use_agents_leverage_real_world_saas_to_solve_professional_workflows/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"SafeArena":{"url":"https://benchmarklist.com/benchmarks/safearena/","metric":"Normalized Safety Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Sahabat-AI - Indonesian (ID)":{"url":"https://huggingface.co/spaces/Sahabat-AI/Sahabat-AI-Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Sahabat-AI - Indonesian Average":{"url":"https://huggingface.co/spaces/Sahabat-AI/Sahabat-AI-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Sahabat-AI - Javanese (JV)":{"url":"https://huggingface.co/spaces/Sahabat-AI/Sahabat-AI-Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Sahabat-AI - Sundanese (SU)":{"url":"https://huggingface.co/spaces/Sahabat-AI/Sahabat-AI-Leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"SalesBench":{"url":"https://hamzamostafa.com/blog/salesbench","metric":"50-Lead Reward","theoretical_min":null,"theoretical_max":null},"Sandboxed Coding Agents are Competitive Omni-m":{"url":"https://benchmarklist.com/benchmarks/sandboxed_coding_agents_are_competitive_omni_modal_task_solvers/","metric":"OmniGAIA Avg. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Scammer4U":{"url":"https://benchmarklist.com/benchmarks/scammer4u/","metric":"C0 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"SceneFunRI":{"url":"https://benchmarklist.com/benchmarks/scenefunri_reasoning_the_invisible_for_task_driven_functional_object_localization/","metric":"cacc_75 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"SciArena":{"url":"https://sciarena.allen.ai/","metric":"Elo Rating","theoretical_min":null,"theoretical_max":null},"SciCode":{"url":"https://scicode-bench.github.io/leaderboard/","metric":"Subproblem Resolve Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"SciDiagramEdit - Aesthetic Preference":{"url":"https://arxiv.org/html/2607.15272v1","metric":"No-skill pairwise win rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"SciDiagramEdit - Semantic Faithfulness":{"url":"https://arxiv.org/html/2607.15272v1","metric":"No-skill pairwise win rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"SciEval - Astronomy":{"url":"https://huggingface.co/spaces/InternScience/SciEval-Leaderboard","metric":"Astronomy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SciEval - Chemistry":{"url":"https://huggingface.co/spaces/InternScience/SciEval-Leaderboard","metric":"Chemistry (%)","theoretical_min":0.0,"theoretical_max":100.0},"SciEval - Earth Sciences":{"url":"https://huggingface.co/spaces/InternScience/SciEval-Leaderboard","metric":"Earth Sciences (%)","theoretical_min":0.0,"theoretical_max":100.0},"SciEval - Life Sciences":{"url":"https://huggingface.co/spaces/InternScience/SciEval-Leaderboard","metric":"Life Sciences (%)","theoretical_min":0.0,"theoretical_max":100.0},"SciEval - Materials Science":{"url":"https://huggingface.co/spaces/InternScience/SciEval-Leaderboard","metric":"Materials Science (%)","theoretical_min":0.0,"theoretical_max":100.0},"SciEval - Overall":{"url":"https://huggingface.co/spaces/InternScience/SciEval-Leaderboard","metric":"Overall (%)","theoretical_min":0.0,"theoretical_max":100.0},"SciEval - Physics":{"url":"https://huggingface.co/spaces/InternScience/SciEval-Leaderboard","metric":"Physics (%)","theoretical_min":0.0,"theoretical_max":100.0},"SciEvalKit":{"url":"https://github.com/InternScience/SciEvalKit","metric":"Scientific Capability Score","theoretical_min":0.0,"theoretical_max":100.0},"SciPaths":{"url":"https://benchmarklist.com/benchmarks/scipaths_forecasting_pathways_to_scientific_discovery/","metric":"F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"SciPredict":{"url":"https://benchmarklist.com/benchmarks/scipredict/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Science Leaderboard":{"url":"https://huggingface.co/spaces/wenhu/Science-Leaderboard","metric":"Average Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"ScratchWorld - Compute":{"url":"https://huggingface.co/spaces/astarforbae/ScratchWorld-Leaderboard","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"ScratchWorld - Create":{"url":"https://huggingface.co/spaces/astarforbae/ScratchWorld-Leaderboard","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"ScratchWorld - Debug":{"url":"https://huggingface.co/spaces/astarforbae/ScratchWorld-Leaderboard","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"ScratchWorld - Extend":{"url":"https://huggingface.co/spaces/astarforbae/ScratchWorld-Leaderboard","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"ScratchWorld - Overall":{"url":"https://huggingface.co/spaces/astarforbae/ScratchWorld-Leaderboard","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"ScreenSpot-Pro":{"url":"https://benchmarklist.com/benchmarks/screenspot_pro/","metric":"Grounding score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ScreenSpot-Pro (Anthropic No Tools)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"ScreenSpot-Pro (Anthropic Tools)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"ScreenSpot-Pro (Mythos Preview No Tools)":{"url":"https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"ScreenSpot-Pro (Mythos Preview Tools)":{"url":"https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Cross-Lingual Consistency - Cross-LogiQA (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Cross-Lingual Consistency - Cross-LogiQA (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Cross-Lingual Consistency - Cross-MMLU (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Cross-Lingual Consistency - Cross-MMLU (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Cross-Lingual Consistency - Cross-XQuAD (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Cross-Lingual Consistency - Cross-XQuAD (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Cultural Reasoning - CN-Eval (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Cultural Reasoning - CN-Eval (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Cultural Reasoning - PH-Eval (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Cultural Reasoning - PH-Eval (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Cultural Reasoning - SG-Eval (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Cultural Reasoning - SG-Eval (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Cultural Reasoning - SG-Eval v1 Cleaned (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Cultural Reasoning - SG-Eval v1 Cleaned (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Cultural Reasoning - SG-Eval v2 MCQ (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Cultural Reasoning - SG-Eval v2 Open (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Cultural Reasoning - US-Eval (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Cultural Reasoning - US-Eval (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Dialogue - DREAM (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Dialogue - DREAM (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Dialogue - DialogSum (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Average ROUGE (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Dialogue - SAMSum (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Average ROUGE (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Emotion - IndoEmotion (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Emotion - IndoEmotion (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Emotion - SST-2 (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Emotion - SST-2 (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - FLORES Translation - Chinese-to-English (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"BLEU (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - FLORES Translation - Chinese-to-English (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"BLEU (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - FLORES Translation - Indonesian-to-English (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"BLEU (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - FLORES Translation - Indonesian-to-English (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"BLEU (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - FLORES Translation - Malay-to-English (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"BLEU (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - FLORES Translation - Malay-to-English (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"BLEU (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - FLORES Translation - Vietnamese-to-English (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"BLEU (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - FLORES Translation - Vietnamese-to-English (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"BLEU (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Fundamental NLP Tasks - C3 (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Fundamental NLP Tasks - C3 (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Fundamental NLP Tasks - CoLA (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Fundamental NLP Tasks - CoLA (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Fundamental NLP Tasks - MNLI (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Fundamental NLP Tasks - MNLI (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Fundamental NLP Tasks - MRPC (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Fundamental NLP Tasks - MRPC (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Fundamental NLP Tasks - OCNLI (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Fundamental NLP Tasks - OCNLI (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Fundamental NLP Tasks - QNLI (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Fundamental NLP Tasks - QNLI (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Fundamental NLP Tasks - QQP (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Fundamental NLP Tasks - QQP (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Fundamental NLP Tasks - RTE (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Fundamental NLP Tasks - RTE (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Fundamental NLP Tasks - WNLI (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Fundamental NLP Tasks - WNLI (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Multilingual Reasoning - C-Eval (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Multilingual Reasoning - C-Eval (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Multilingual Reasoning - CMMLU (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Multilingual Reasoning - CMMLU (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Multilingual Reasoning - IndoMMLU (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Multilingual Reasoning - IndoMMLU (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Multilingual Reasoning - MMLU (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Multilingual Reasoning - MMLU (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Multilingual Reasoning - ZBench (Few-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeaEval - Multilingual Reasoning - ZBench (Zero-Shot)":{"url":"https://huggingface.co/spaces/MERaLiON/SeaEval_Leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SearchGEO":{"url":"https://benchmarklist.com/benchmarks/searchgeo/","metric":"ASR↓ (%) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"SecCodeBench":{"url":"https://alibaba.github.io/sec-code-bench/","metric":"Total Score","theoretical_min":0.0,"theoretical_max":100.0},"SecCodePLT":{"url":"https://huggingface.co/spaces/FrontierAICybersecurity/Cybersecurity_leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeePhys":{"url":"https://seephys.github.io/","metric":"Total Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SeePhys Pro":{"url":"https://benchmarklist.com/benchmarks/seephys_pro_diagnosing_modality_transfer_and_blind_training_effects_in_multimodal_rlvr_for_physics_reasoning/","metric":"Cons4 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Seeing Culture - Grounding mIoU":{"url":"https://seeingculture-benchmark.github.io/","metric":"Overall mIoU (%)","theoretical_min":0.0,"theoretical_max":100.0},"Seeing Culture - VQA Accuracy":{"url":"https://seeingculture-benchmark.github.io/","metric":"Overall Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Seeing Isn't Knowing":{"url":"https://benchmarklist.com/benchmarks/seeing_isn_t_knowing_do_vlms_know_when_not_to_answer_spatial_questions_and_why/","metric":"Occlusion All (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Self-Driving Negotiator":{"url":"https://arxiv.org/html/2606.15139v1","metric":"Average success rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Self-Ensembling Vision-Language Models for Cha":{"url":"https://benchmarklist.com/benchmarks/self_ensembling_vision_language_models_for_chart_data_extraction/","metric":"ChartQA (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Semgrep IDOR - Guided Prompt F1":{"url":"https://semgrep.dev/blog/2026/gpt-5-6-benchmarks-ai-code-security/","metric":"F1 score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Semgrep IDOR - Multimodal F1":{"url":"https://semgrep.dev/blog/2026/gpt-5-6-benchmarks-ai-code-security/","metric":"F1 score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Seneca-TRBench":{"url":"https://benchmarklist.com/benchmarks/seneca_trbench/","metric":"Combined Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Senior SWE-Bench":{"url":"https://senior-swe-bench.snorkel.ai/","metric":"Tasteful Solve Rate (pass@1, %)","theoretical_min":0.0,"theoretical_max":100.0},"Sense Ruby Navigation - Baseline":{"url":"https://github.com/luuuc/sense/tree/main/bench/verticals/ruby-rails/results","metric":"Mean cited recall across repositories (%)","theoretical_min":0.0,"theoretical_max":100.0},"Shader Benchmark":{"url":"https://nbardy.github.io/shader_benchmark/","metric":"Avg judge score (0-500, render fails count 0)","theoretical_min":0.0,"theoretical_max":500.0},"ShaderMatch":{"url":"https://huggingface.co/spaces/Vipitis/shadermatch","metric":"Clone Match Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Shlepa - Books MC (RU)":{"url":"https://huggingface.co/spaces/Vikhrmodels/small-shlepa-lb","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Shlepa - Law MC (RU)":{"url":"https://huggingface.co/spaces/Vikhrmodels/small-shlepa-lb","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Shlepa - Movies MC (RU)":{"url":"https://huggingface.co/spaces/Vikhrmodels/small-shlepa-lb","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Shlepa - Music MC (RU)":{"url":"https://huggingface.co/spaces/Vikhrmodels/small-shlepa-lb","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SimpleBench":{"url":"https://simple-bench.com/","metric":"Score (AVG@5)","theoretical_min":0.0,"theoretical_max":100.0},"SimpleQA Verified":{"url":"https://epoch.ai/benchmarks/simple-qa-verified","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SkateBench":{"url":"https://skatebench.t3.gg/","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"SkillsBench":{"url":"https://www.skillsbench.ai","metric":"With-skills score (%)","theoretical_min":0.0,"theoretical_max":100.0},"SlopCodeBench":{"url":"https://www.scbench.ai/leaderboard","metric":"Isolated Solved (%)","theoretical_min":0.0,"theoretical_max":100.0},"Slovenian OCR Benchmark":{"url":"https://huggingface.co/spaces/valira-ai/slo-ocr-benchmark","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"SmartHome-Bench":{"url":"https://benchmarklist.com/benchmarks/smarthome_bench/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"SmellBench":{"url":"https://benchmarklist.com/benchmarks/smellbench_evaluating_llm_agents_on_architectural_code_smell_repair/","metric":"Weighted Effectiveness (E) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Smolagents - GAIA":{"url":"https://huggingface.co/spaces/smolagents/smolagents-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Smolagents - MATH":{"url":"https://huggingface.co/spaces/smolagents/smolagents-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Smolagents - SimpleQA":{"url":"https://huggingface.co/spaces/smolagents/smolagents-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Smolagents Leaderboard":{"url":"https://huggingface.co/spaces/smolagents/smolagents-leaderboard","metric":"Average Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SnakeBench":{"url":"https://snakebench.com/","metric":"TrueSkill Rating","theoretical_min":0.0,"theoretical_max":100.0},"SnitchBench":{"url":"https://snitchbench.t3.gg/","metric":"Govt Contact Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"SocialPersona":{"url":"https://benchmarklist.com/benchmarks/socialpersona_benchmarking_personalized_profiling_and_response_with_multimodal_social_media_context/","metric":"Int. F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Software Engineering Arena - Model Arena":{"url":"https://huggingface.co/spaces/SE-Arena/Software-Engineering-Arena","metric":"Elo Rating","theoretical_min":null,"theoretical_max":null},"SolarBench":{"url":"https://solarbench.maingen.ai/","metric":"Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Soohak":{"url":"https://benchmarklist.com/benchmarks/soohak/","metric":"Pass@3 (Challenge) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"SpaceDG":{"url":"https://benchmarklist.com/benchmarks/spacedg/","metric":"Avg. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"SpaceVista":{"url":"https://peiwensun2000.github.io/mm2km/","metric":"Overall Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"SpacetimeDB LLM Benchmark (C#)":{"url":"https://spacetimedb.com/llms-benchmark?lang=csharp","metric":"Task Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"SpacetimeDB LLM Benchmark (Rust)":{"url":"https://spacetimedb.com/llms-benchmark?lang=rust","metric":"Task Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"SpacetimeDB LLM Benchmark (TypeScript)":{"url":"https://spacetimedb.com/llms-benchmark?lang=typescript","metric":"Task Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"SpatialAct":{"url":"https://benchmarklist.com/benchmarks/spatialact/","metric":"Succ. Rate (self-reported)","theoretical_min":0.0,"theoretical_max":null},"SpatialBabel":{"url":"https://benchmarklist.com/benchmarks/spatialbabel/","metric":"Three.js (self-reported)","theoretical_min":0.0,"theoretical_max":null},"SpeechMap Compliance":{"url":"https://speechmap.ai/models/","metric":"% Requests Completed","theoretical_min":0.0,"theoretical_max":100.0},"Spider 2.0-DBT":{"url":"https://spider2-sql.github.io/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Spider 2.0-Lite":{"url":"https://spider2-sql.github.io/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Spider 2.0-Snow":{"url":"https://spider2-sql.github.io/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Spiral-Bench (Delusion)":{"url":"https://eqbench.com/spiral-bench.html","metric":"Delusion Score (0-100)","theoretical_min":0.0,"theoretical_max":100.0},"SpreadsheetBench":{"url":"https://spreadsheetbench.github.io/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"StakeBench":{"url":"https://benchmarklist.com/benchmarks/stakebench/","metric":"Agg (self-reported)","theoretical_min":0.0,"theoretical_max":null},"StatABench":{"url":"https://benchmarklist.com/benchmarks/statabench_dataset_and_framework_for_evaluating_statistical_analysis_capabilities_of_llms/","metric":"Total (self-reported)","theoretical_min":0.0,"theoretical_max":null},"StemBind":{"url":"https://benchmarklist.com/benchmarks/stembind_when_mllms_get_lost_between_rules_and_instances_in_abstract_visual_reasoning/","metric":"F Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Step Game (Lechmazur)":{"url":"https://github.com/lechmazur/step_game","metric":"TrueSkill μ","theoretical_min":null,"theoretical_max":null},"StereoSet":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"ICAT Score","theoretical_min":0.0,"theoretical_max":100.0},"StereoTales":{"url":"https://benchmarklist.com/benchmarks/stereotales/","metric":"Emissions (self-reported)","theoretical_min":0.0,"theoretical_max":null},"StickToYourRole":{"url":"https://huggingface.co/spaces/flowers-team/StickToYourRoleLeaderboard","metric":"Cardinal Score","theoretical_min":0.0,"theoretical_max":100.0},"Story Theory Bench":{"url":"https://github.com/clchinkc/story-bench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"StoryAlign":{"url":"https://benchmarklist.com/benchmarks/storyalign_evaluating_and_training_reward_models_for_story_generation/","metric":"Average (self-reported)","theoretical_min":0.0,"theoretical_max":null},"StreamProfileBench":{"url":"https://benchmarklist.com/benchmarks/streamprofilebench_a_benchmark_for_fine_grained_user_profile_inference_in_real_world_streaming_scenarios/","metric":"F1_NS Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"StressEval":{"url":"https://benchmarklist.com/benchmarks/stresseval_failure_driven_dynamic_benchmarking_for_knowledge_intensive_reasoning_in_large_language_models/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"StructEval":{"url":"https://benchmarklist.com/benchmarks/structeval/","metric":"Average (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Structural Biology (Anthropic)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Structural Biology Multiple Choice (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Structural Biology Open-Ended (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"StructureClaw (Automatic Workflow)":{"url":"https://arxiv.org/abs/2607.14896","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Structured Output Benchmark":{"url":"https://interfaze.ai/leaderboards/structured-output-benchmark","metric":"Overall (%)","theoretical_min":0.0,"theoretical_max":100.0},"SudokuBench (Multi-step, 4x4)":{"url":"https://pub.sakana.ai/sudoku/","metric":"4x4 Solve Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"SudokuBench (Multi-step, 6x6)":{"url":"https://pub.sakana.ai/sudoku/","metric":"6x6 Solve Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"SudokuBench (Multi-step, 9x9)":{"url":"https://pub.sakana.ai/sudoku/","metric":"9x9 Solve Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"SudokuBench (Single-shot, 4x4)":{"url":"https://pub.sakana.ai/sudoku/","metric":"4x4 Solve Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"SudokuBench (Single-shot, 6x6)":{"url":"https://pub.sakana.ai/sudoku/","metric":"6x6 Solve Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"SudokuBench (Single-shot, 9x9)":{"url":"https://pub.sakana.ai/sudoku/","metric":"9x9 Solve Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"SuiChat-CN":{"url":"https://benchmarklist.com/benchmarks/suichat_cn_benchmarking_contextual_suicide_risk_assessment_in_chinese_group_chats/","metric":"Standard F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"SuperCLUE":{"url":"https://www.cluebenchmarks.com/superclue.html","metric":"Total Score","theoretical_min":0.0,"theoretical_max":100.0},"SuperGPQA":{"url":"https://supergpqa.github.io/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":10},"Surface Evolver Bench":{"url":"https://yhenon.github.io/surface-evolver-llm-eval/","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Surface Evolver Bench Pass Rate":{"url":"https://yhenon.github.io/surface-evolver-llm-eval/","metric":"Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"SvelteBench":{"url":"https://khromov.github.io/svelte-bench/benchmark-results-merged.html","metric":"Average pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"Sycophancy (Lechmazur)":{"url":"https://github.com/lechmazur/sycophancy","metric":"Sycophancy rate % (lower is better)","theoretical_min":0.0,"theoretical_max":100.0},"SynthPAI":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"Average accuracy in %","theoretical_min":0.0,"theoretical_max":100.0},"T-Eval":{"url":"https://open-compass.github.io/T-Eval/leaderboard.html","metric":"Overall Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"TACTL":{"url":"https://huggingface.co/spaces/FrontierAICybersecurity/Cybersecurity_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"TAEF (Text Analytics Evaluation Framework)":{"url":"https://benchmarklist.com/benchmarks/taef_text_analytics_evaluation_framework/","metric":"F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TERMS-Bench":{"url":"https://terms-bench.github.io/","metric":"Mean Utility","theoretical_min":null,"theoretical_max":null},"TIDE-Bench":{"url":"https://benchmarklist.com/benchmarks/tide_bench_task_aware_and_diagnostic_evaluation_of_tool_integrated_reasoning/","metric":"GSM8K Acc. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TIR-Bench":{"url":"https://benchmarklist.com/benchmarks/tir_bench/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TOBench":{"url":"https://benchmarklist.com/benchmarks/tobench_a_task_oriented_omni_modal_benchmark_for_real_world_tool_using_agents/","metric":"Avg. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TRACER":{"url":"https://benchmarklist.com/benchmarks/tracer_a_semantic_aware_framework_for_fine_grained_contamination_detection_in_code_llms/","metric":"F1 Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TRAIL GAIA":{"url":"https://huggingface.co/spaces/PatronusAI/TRAIL","metric":"Joint Accuracy","theoretical_min":0.0,"theoretical_max":100.0},"TRAIL SWE":{"url":"https://huggingface.co/spaces/PatronusAI/TRAIL","metric":"Joint Accuracy","theoretical_min":0.0,"theoretical_max":100.0},"TRAP":{"url":"https://benchmarklist.com/benchmarks/trap_benchmark_for_task_completion_and_resistance_to_active_privacy_extraction/","metric":"Overall HM (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TRLawBench":{"url":"https://benchmarklist.com/benchmarks/trlawbench/","metric":"Stage 1 accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TSCG":{"url":"https://benchmarklist.com/benchmarks/tscg_deterministic_tool_schema_compilation_for_agentic_llm_deployments/","metric":"20 Tools (json-text) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TVBench":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"Average Accuracy","theoretical_min":0.0,"theoretical_max":100.0},"TW-LegalBench":{"url":"https://benchmarklist.com/benchmarks/tw_legalbench_measuring_taiwanese_legal_understanding/","metric":"Total (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TableBench":{"url":"https://tablebench.github.io/","metric":"DP Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"TableVista":{"url":"https://benchmarklist.com/benchmarks/tablevista_benchmarking_multimodal_table_reasoning_under_visual_and_structural_complexity/","metric":"Avg. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Tau-Bench Airline":{"url":"https://taubench.com/","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"Tau-Bench Retail":{"url":"https://taubench.com/","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"Tau-Bench Telecom":{"url":"https://taubench.com/","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"Tau2 Bench Retail (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Tau2 Bench Telecom (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Tau3 Banking":{"url":"https://artificialanalysis.ai/evaluations/tau3-banking","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Tau3-Bench Airline":{"url":"https://taubench.com/","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"Tau3-Bench Banking_Knowledge":{"url":"https://taubench.com/","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"Tau3-Bench Retail":{"url":"https://taubench.com/","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"Tau3-Bench Telecom":{"url":"https://taubench.com/","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"TaxBench":{"url":"https://benchmarklist.com/benchmarks/taxbench/","metric":"Mean pass^5 (computed) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TaxCalcBench":{"url":"https://github.com/column-tax/tax-calc-bench","metric":"Correct Returns - Strict (%)","theoretical_min":0.0,"theoretical_max":100.0},"TeamBench":{"url":"https://benchmarklist.com/benchmarks/teambench_evaluating_agent_coordination_under_enforced_role_separation/","metric":"Solo (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TeleResilienceBench":{"url":"https://benchmarklist.com/benchmarks/teleresiliencebench_quantifying_resilience_for_llm_reasoning_in_telecommunications/","metric":"Macro Average CFR (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TempCompass":{"url":"https://huggingface.co/spaces/lyx97/TempCompass","metric":"Avg All (%)","theoretical_min":0.0,"theoretical_max":100.0},"TempGlitch":{"url":"https://benchmarklist.com/benchmarks/tempglitch_evaluating_vision_language_models_for_temporal_glitch_detection_in_gameplay_videos/","metric":"Acc. (1 FPS) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Terminal Bench":{"url":"https://benchmarklist.com/benchmarks/terminal_bench_epoch/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Terminal-Bench 2.0":{"url":"https://www.tbench.ai/leaderboard/terminal-bench/2.0","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Terminal-Bench 2.0 (Mythos Preview System Card)":{"url":"https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf","metric":"Mean Reward (%)","theoretical_min":0.0,"theoretical_max":100.0},"Terminal-Bench 2.0 (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Terminal-Bench 2.1":{"url":"https://www.tbench.ai/leaderboard/terminal-bench/2.1","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Terminal-Bench 2.1 (Claude Code)":{"url":"https://www.tbench.ai/leaderboard/terminal-bench/2.1","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Terminal-Bench 2.1 (Fable/Mythos)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Mean Reward (%)","theoretical_min":0.0,"theoretical_max":100.0},"TerminalWorld":{"url":"https://terminalworld.ai/leaderboard/","metric":"Pass Rate (%, 200 verified tasks)","theoretical_min":0.0,"theoretical_max":null},"TextClass Benchmark":{"url":"https://benchmarklist.com/benchmarks/textclass_benchmark/","metric":"Meta-Elo (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TextQuests (No Clues)":{"url":"https://www.textquests.ai/","metric":"Avg Game Progress (%)","theoretical_min":0.0,"theoretical_max":100.0},"TextQuests (With Clues)":{"url":"https://www.textquests.ai/","metric":"Avg Game Progress (%)","theoretical_min":0.0,"theoretical_max":100.0},"ThaiSafetyBench":{"url":"https://benchmarklist.com/benchmarks/thaisafetybench/","metric":"Overall ASR (self-reported)","theoretical_min":0.0,"theoretical_max":null},"The Age of Curiosity Meets the Age of AI: Benc":{"url":"https://benchmarklist.com/benchmarks/the_age_of_curiosity_meets_the_age_of_ai_benchmarking_child_safety_in_large_language_models/","metric":"Total (self-reported)","theoretical_min":0.0,"theoretical_max":null},"The Compliance Gap":{"url":"https://benchmarklist.com/benchmarks/the_compliance_gap_why_ai_systems_promise_to_follow_process_instructions_but_don_t/","metric":"Authority (self-reported)","theoretical_min":0.0,"theoretical_max":null},"The Compliance Trap":{"url":"https://benchmarklist.com/benchmarks/the_compliance_trap_how_structural_constraints_degrade_frontier_ai_metacognition_under_adversarial_pressure/","metric":"Δ (degradation) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"The Metanym Game":{"url":"https://benchmarklist.com/benchmarks/the_metanym_game_a_self_contained_self_consistent_llm_peer_community_benchmark_for_structural_intelligence/","metric":"T (self-reported)","theoretical_min":0.0,"theoretical_max":null},"The Wrong Kind of Right":{"url":"https://benchmarklist.com/benchmarks/the_wrong_kind_of_right_quantifying_and_localizing_misfired_alignment_in_llms/","metric":"MAR Disability (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Time to REFLECT":{"url":"https://benchmarklist.com/benchmarks/time_to_reflect_can_we_trust_llm_judges_for_evidence_based_research_agents/","metric":"Overall (Report Quality) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TimeSage-MT":{"url":"https://benchmarklist.com/benchmarks/timesage_mt/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TinyQABenchmark (core-en)":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"Exact Match","theoretical_min":0.0,"theoretical_max":100.0},"ToolBench - Google Sheets":{"url":"https://huggingface.co/spaces/qiantong-xu/toolbench-leaderboard","metric":"Task Score","theoretical_min":0.0,"theoretical_max":100.0},"ToolBench - Home Search":{"url":"https://huggingface.co/spaces/qiantong-xu/toolbench-leaderboard","metric":"Task Score","theoretical_min":0.0,"theoretical_max":100.0},"ToolBench - Open Weather":{"url":"https://huggingface.co/spaces/qiantong-xu/toolbench-leaderboard","metric":"Task Score","theoretical_min":0.0,"theoretical_max":100.0},"ToolBench - Tabletop":{"url":"https://huggingface.co/spaces/qiantong-xu/toolbench-leaderboard","metric":"Task Score","theoretical_min":0.0,"theoretical_max":100.0},"ToolBench - The Cat API":{"url":"https://huggingface.co/spaces/qiantong-xu/toolbench-leaderboard","metric":"Task Score","theoretical_min":0.0,"theoretical_max":100.0},"ToolBench - Trip Booking":{"url":"https://huggingface.co/spaces/qiantong-xu/toolbench-leaderboard","metric":"Task Score","theoretical_min":0.0,"theoretical_max":100.0},"ToolBench - VirtualHome":{"url":"https://huggingface.co/spaces/qiantong-xu/toolbench-leaderboard","metric":"Task Score","theoretical_min":0.0,"theoretical_max":100.0},"ToolBench - WebShop Long":{"url":"https://huggingface.co/spaces/qiantong-xu/toolbench-leaderboard","metric":"Task Score","theoretical_min":0.0,"theoretical_max":100.0},"ToolBench - WebShop Short":{"url":"https://huggingface.co/spaces/qiantong-xu/toolbench-leaderboard","metric":"Task Score","theoretical_min":0.0,"theoretical_max":100.0},"ToolBench Leaderboard":{"url":"https://huggingface.co/spaces/qiantong-xu/toolbench-leaderboard","metric":"Average Task Score","theoretical_min":0.0,"theoretical_max":100.0},"ToolSandbox":{"url":"https://benchmarklist.com/benchmarks/toolsandbox/","metric":"Avg Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Toolathlon":{"url":"https://benchmarklist.com/benchmarks/toolathlon/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Toolathlon (Anthropic Internal Harness)":{"url":"https://www-cdn.anthropic.com/d00db56fa754a1b115b6dd7cb2e3c342ee809620.pdf","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"TraceEval":{"url":"https://benchmarklist.com/benchmarks/traceeval/","metric":"Average F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TrackingAI IQ Test":{"url":"https://trackingai.org/","metric":"IQ Test Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"TrackingAI IQ Test (Mensa Norway)":{"url":"https://trackingai.org/","metric":"Mensa Norway Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"TrackingAI IQ Test (Offline)":{"url":"https://trackingai.org/","metric":"Offline IQ Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"TrackingAI IQ Test (Vision)":{"url":"https://trackingai.org/","metric":"IQ Test Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Translation (Lechmazur)":{"url":"https://github.com/lechmazur/translation","metric":"Mean Score","theoretical_min":1.0,"theoretical_max":10.0},"Translation Set1 to en COMET22":{"url":"https://benchmarklist.com/benchmarks/translation_set1_en_comet22/","metric":"Set1→en COMET22 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Translation Set1 to en spBleu":{"url":"https://benchmarklist.com/benchmarks/translation_set1_en_spbleu/","metric":"Set1→en spBLEU (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Translation en to Set1 COMET22":{"url":"https://benchmarklist.com/benchmarks/translation_en_set1_comet22/","metric":"en→Set1 COMET22 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Translation en to Set1 spBleu":{"url":"https://benchmarklist.com/benchmarks/translation_en_set1_spbleu/","metric":"en→Set1 spBLEU (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TriBench-Ko":{"url":"https://benchmarklist.com/benchmarks/tribench_ko/","metric":"Macro F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TriViewBench":{"url":"https://benchmarklist.com/benchmarks/triviewbench_controlled_complexity_scaling_for_multi_view_structural_reasoning_in_mllms/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TriggerBench":{"url":"https://benchmarklist.com/benchmarks/triggerbench/","metric":"Slot Match (Positive Clean) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Trip+":{"url":"https://benchmarklist.com/benchmarks/trip/","metric":"Plan Avg. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TriviaQA":{"url":"https://epoch.ai/benchmarks/eci","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Truthful AI Advisors (Crawford-Sobel cheap-tal":{"url":"https://benchmarklist.com/benchmarks/truthful_ai_advisors_crawford_sobel_cheap_talk_benchmark/","metric":"Mean Normalized MI (positive bias) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TuRTLe - Icarus Area":{"url":"https://huggingface.co/spaces/HPAI-BSC/TuRTLe-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"TuRTLe - Icarus EM":{"url":"https://huggingface.co/spaces/HPAI-BSC/TuRTLe-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"TuRTLe - Icarus Functionality":{"url":"https://huggingface.co/spaces/HPAI-BSC/TuRTLe-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"TuRTLe - Icarus Performance":{"url":"https://huggingface.co/spaces/HPAI-BSC/TuRTLe-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"TuRTLe - Icarus Power":{"url":"https://huggingface.co/spaces/HPAI-BSC/TuRTLe-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"TuRTLe - Icarus Syntax":{"url":"https://huggingface.co/spaces/HPAI-BSC/TuRTLe-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"TuRTLe - Icarus Synthesis":{"url":"https://huggingface.co/spaces/HPAI-BSC/TuRTLe-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"TuRTLe - Verilator Area":{"url":"https://huggingface.co/spaces/HPAI-BSC/TuRTLe-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"TuRTLe - Verilator EM":{"url":"https://huggingface.co/spaces/HPAI-BSC/TuRTLe-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"TuRTLe - Verilator Functionality":{"url":"https://huggingface.co/spaces/HPAI-BSC/TuRTLe-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"TuRTLe - Verilator Performance":{"url":"https://huggingface.co/spaces/HPAI-BSC/TuRTLe-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"TuRTLe - Verilator Power":{"url":"https://huggingface.co/spaces/HPAI-BSC/TuRTLe-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"TuRTLe - Verilator Syntax":{"url":"https://huggingface.co/spaces/HPAI-BSC/TuRTLe-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"TuRTLe - Verilator Synthesis":{"url":"https://huggingface.co/spaces/HPAI-BSC/TuRTLe-Leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"TuRTLe Code Completion (Icarus Verilog)":{"url":"https://benchmarklist.com/benchmarks/turtle_code_completion_icarus/","metric":"Aggregated Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TuRTLe Code Completion (Verilator)":{"url":"https://benchmarklist.com/benchmarks/turtle_code_completion_verilator/","metric":"Aggregated Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TuRTLe Module Completion (NotSoTiny)":{"url":"https://benchmarklist.com/benchmarks/turtle_module_completion_not_so_tiny/","metric":"Functionality EQV (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TuRTLe Spec-to-RTL (Icarus Verilog)":{"url":"https://benchmarklist.com/benchmarks/turtle_spec_to_rtl_icarus/","metric":"Aggregated Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TuRTLe Spec-to-RTL (Verilator)":{"url":"https://benchmarklist.com/benchmarks/turtle_spec_to_rtl_verilator/","metric":"Aggregated Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"TukaBench":{"url":"https://benchmarklist.com/benchmarks/tukabench_a_culturally_grounded_jailbreak_benchmark_for_african_languages/","metric":"ASR ↑ African Languages (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Turkish MMLU":{"url":"https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"TutorBench":{"url":"https://benchmarklist.com/benchmarks/tutorbench/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"U-MATH":{"url":"https://huggingface.co/spaces/toloka/u-math-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"U-MATH - Algebra":{"url":"https://huggingface.co/spaces/toloka/u-math-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"U-MATH - Differential Calculus":{"url":"https://huggingface.co/spaces/toloka/u-math-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"U-MATH - Integral Calculus":{"url":"https://huggingface.co/spaces/toloka/u-math-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"U-MATH - Multivariable Calculus":{"url":"https://huggingface.co/spaces/toloka/u-math-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"U-MATH - Precalculus":{"url":"https://huggingface.co/spaces/toloka/u-math-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"U-MATH - Sequences & Series":{"url":"https://huggingface.co/spaces/toloka/u-math-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"UGI - Natural Intelligence":{"url":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","metric":"NatInt Score","theoretical_min":0.0,"theoretical_max":100.0},"UGI - Willingness (W/10)":{"url":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","metric":"W/10 Score","theoretical_min":0.0,"theoretical_max":10.0},"UGI - Writing":{"url":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","metric":"Writing Score","theoretical_min":0.0,"theoretical_max":100.0},"UGI Leaderboard":{"url":"https://huggingface.co/spaces/DontPlanToEnd/UGI-Leaderboard","metric":"UGI Score","theoretical_min":0.0,"theoretical_max":100.0},"URIAL-Bench - Coding":{"url":"https://huggingface.co/spaces/allenai/URIAL-Bench","metric":"Judge Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"URIAL-Bench - Extraction":{"url":"https://huggingface.co/spaces/allenai/URIAL-Bench","metric":"Judge Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"URIAL-Bench - Humanities":{"url":"https://huggingface.co/spaces/allenai/URIAL-Bench","metric":"Judge Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"URIAL-Bench - Math":{"url":"https://huggingface.co/spaces/allenai/URIAL-Bench","metric":"Judge Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"URIAL-Bench - Overall":{"url":"https://huggingface.co/spaces/allenai/URIAL-Bench","metric":"Judge Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"URIAL-Bench - Reasoning":{"url":"https://huggingface.co/spaces/allenai/URIAL-Bench","metric":"Judge Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"URIAL-Bench - Roleplay":{"url":"https://huggingface.co/spaces/allenai/URIAL-Bench","metric":"Judge Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"URIAL-Bench - STEM":{"url":"https://huggingface.co/spaces/allenai/URIAL-Bench","metric":"Judge Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"URIAL-Bench - Turn 1":{"url":"https://huggingface.co/spaces/allenai/URIAL-Bench","metric":"Judge Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"URIAL-Bench - Turn 2":{"url":"https://huggingface.co/spaces/allenai/URIAL-Bench","metric":"Judge Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"URIAL-Bench - Writing":{"url":"https://huggingface.co/spaces/allenai/URIAL-Bench","metric":"Judge Score (0-10)","theoretical_min":0.0,"theoretical_max":10.0},"USAMO 2026 (Anthropic)":{"url":"https://www-cdn.anthropic.com/0b4915911bb0d19eca5b5ee635c80fef830a37ea/Claude%20Opus%204.8%20System%20Card.pdf","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"USAMO 2026 (Mythos Preview System Card)":{"url":"https://www-cdn.anthropic.com/8b8380204f74670be75e81c820ca8dda846ab289.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"USAMO25":{"url":"https://benchmarklist.com/benchmarks/usamo25/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"UX Leaderboard":{"url":"https://huggingface.co/spaces/nlpetprolific/user-experience-leaderboard","metric":"Overall Score","theoretical_min":0.0,"theoretical_max":100.0},"UXBench":{"url":"https://benchmarklist.com/benchmarks/uxbench/","metric":"Automated Lift (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Ukrainian LLM - ARC Challenge UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - ARC Easy UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Belebele UKR Cyrl":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Flores CRH UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Flores CS UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Flores DE UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Flores EN UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Flores HU UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Flores PL UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Flores RO UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Flores RU UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Flores SK UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Flores UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Flores UK CRH":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Flores UK CS":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Flores UK DE":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Flores UK EN":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Flores UK HU":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Flores UK PL":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Flores UK RO":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Flores UK RU":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Flores UK SK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - GSM8K UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Abstract Algebra":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Anatomy":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Astronomy":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Business Ethics":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Clinical Knowledge":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK College Biology":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK College Chemistry":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK College Computer Science":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK College Mathematics":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK College Medicine":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK College Physics":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Computer Security":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Conceptual Physics":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Econometrics":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Electrical Engineering":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Elementary Mathematics":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Formal Logic":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Global Facts":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK High School Biology":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK High School Chemistry":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK High School Computer Science":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK High School European History":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK High School Geography":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK High School Government AND Politics":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK High School Macroeconomics":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK High School Mathematics":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK High School Microeconomics":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK High School Physics":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK High School Psychology":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK High School Statistics":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK High School US History":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK High School World History":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Human Aging":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Human Sexuality":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Humanities":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK International LAW":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Jurisprudence":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Logical Fallacies":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Machine Learning":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Management":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Marketing":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Medical Genetics":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Miscellaneous":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Moral Disputes":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Moral Scenarios":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Nutrition":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Other":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Philosophy":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Prehistory":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Professional Accounting":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Professional LAW":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Professional Medicine":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Professional Psychology":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Public Relations":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Security Studies":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Social Sciences":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Sociology":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Stem":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK US Foreign Policy":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK Virology":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Global MMLU Full UK World Religions":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - IFEVAL UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Long Flores CRH UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Long Flores CS UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Long Flores DE UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Long Flores EN UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Long Flores HU UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Long Flores PL UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Long Flores RO UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Long Flores RU UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Long Flores SK UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Long Flores UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Long Flores UK CRH":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Long Flores UK CS":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Long Flores UK DE":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Long Flores UK EN":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Long Flores UK HU":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Long Flores UK PL":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Long Flores UK RO":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Long Flores UK RU":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Long Flores UK SK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Squad UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Triviaqa UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - WMT EN UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Winogrande UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - Xlsum UK":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - ZNO UK Geography":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - ZNO UK History":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - ZNO UK Language AND Literature":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM - ZNO UK Math":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian LLM Leaderboard":{"url":"https://huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Ukrainian Legal Text Benchmark (EDRSR subset)":{"url":"https://benchmarklist.com/benchmarks/ukrainian_legal_text_benchmark_edrsr_subset/","metric":"3-task Composite (self-reported)","theoretical_min":0.0,"theoretical_max":null},"UniGenBench":{"url":"https://huggingface.co/spaces/CodeGoat24/UniGenBench_Leaderboard","metric":"Overall Score","theoretical_min":0.0,"theoretical_max":100.0},"UniGenBench - Action":{"url":"https://huggingface.co/spaces/CodeGoat24/UniGenBench_Leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"UniGenBench - Attribute":{"url":"https://huggingface.co/spaces/CodeGoat24/UniGenBench_Leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"UniGenBench - Compound":{"url":"https://huggingface.co/spaces/CodeGoat24/UniGenBench_Leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"UniGenBench - Grammar":{"url":"https://huggingface.co/spaces/CodeGoat24/UniGenBench_Leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"UniGenBench - Layout":{"url":"https://huggingface.co/spaces/CodeGoat24/UniGenBench_Leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"UniGenBench - Relationship":{"url":"https://huggingface.co/spaces/CodeGoat24/UniGenBench_Leaderboard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"UniREditBench":{"url":"https://huggingface.co/spaces/maplebb/UniREditBench_Leaderboard","metric":"Overall Score","theoretical_min":0.0,"theoretical_max":100.0},"UniREditBench - Game World":{"url":"https://huggingface.co/spaces/maplebb/UniREditBench_Leaderboard","metric":"Overall Score","theoretical_min":0.0,"theoretical_max":100.0},"UniREditBench - Real World":{"url":"https://huggingface.co/spaces/maplebb/UniREditBench_Leaderboard","metric":"Overall Score","theoretical_min":0.0,"theoretical_max":100.0},"V-STaR":{"url":"https://huggingface.co/spaces/V-STaR-Bench/V-STaR-LeaderBoard","metric":"Average All Score","theoretical_min":0.0,"theoretical_max":100.0},"V-STaR - All mAM":{"url":"https://huggingface.co/spaces/V-STaR-Bench/V-STaR-LeaderBoard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"V-STaR - All mLGM":{"url":"https://huggingface.co/spaces/V-STaR-Bench/V-STaR-LeaderBoard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"V-STaR - Long mAM":{"url":"https://huggingface.co/spaces/V-STaR-Bench/V-STaR-LeaderBoard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"V-STaR - Long mLGM":{"url":"https://huggingface.co/spaces/V-STaR-Bench/V-STaR-LeaderBoard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"V-STaR - Medium mAM":{"url":"https://huggingface.co/spaces/V-STaR-Bench/V-STaR-LeaderBoard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"V-STaR - Medium mLGM":{"url":"https://huggingface.co/spaces/V-STaR-Bench/V-STaR-LeaderBoard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"V-STaR - Short mAM":{"url":"https://huggingface.co/spaces/V-STaR-Bench/V-STaR-LeaderBoard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"V-STaR - Short mLGM":{"url":"https://huggingface.co/spaces/V-STaR-Bench/V-STaR-LeaderBoard","metric":"Score","theoretical_min":0.0,"theoretical_max":100.0},"VANTAGE-Bench":{"url":"https://huggingface.co/spaces/clemson-computing/VANTAGE-Bench-Leaderboard","metric":"Overall (%)","theoretical_min":0.0,"theoretical_max":100.0},"VANTAGE-Bench - Event Verification":{"url":"https://huggingface.co/spaces/clemson-computing/VANTAGE-Bench-Leaderboard","metric":"Event Verification (%)","theoretical_min":0.0,"theoretical_max":100.0},"VANTAGE-Bench - Semantic":{"url":"https://huggingface.co/spaces/clemson-computing/VANTAGE-Bench-Leaderboard","metric":"Semantic (%)","theoretical_min":0.0,"theoretical_max":100.0},"VANTAGE-Bench - Single Object Tracking":{"url":"https://huggingface.co/spaces/clemson-computing/VANTAGE-Bench-Leaderboard","metric":"Single Object Tracking (%)","theoretical_min":0.0,"theoretical_max":100.0},"VANTAGE-Bench - Spatial":{"url":"https://huggingface.co/spaces/clemson-computing/VANTAGE-Bench-Leaderboard","metric":"Spatial (%)","theoretical_min":0.0,"theoretical_max":100.0},"VANTAGE-Bench - Temporal":{"url":"https://huggingface.co/spaces/clemson-computing/VANTAGE-Bench-Leaderboard","metric":"Temporal (%)","theoretical_min":0.0,"theoretical_max":100.0},"VANTAGE-Bench - Video QA":{"url":"https://huggingface.co/spaces/clemson-computing/VANTAGE-Bench-Leaderboard","metric":"Video QA (%)","theoretical_min":0.0,"theoretical_max":100.0},"VAREX-Bench":{"url":"https://benchmarklist.com/benchmarks/varex_bench/","metric":"Exact Match (self-reported)","theoretical_min":0.0,"theoretical_max":null},"VGBench":{"url":"https://vgbench.github.io/","metric":"Avg Score (SVG+TikZ+Graphviz)","theoretical_min":0.0,"theoretical_max":100.0},"VIBench":{"url":"https://benchmarklist.com/benchmarks/vibench/","metric":"Direct VIB (self-reported)","theoretical_min":null,"theoretical_max":null},"VLM2-Bench":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"Average Score on VLM2-bench (9 subtasks)","theoretical_min":0.0,"theoretical_max":100.0},"VMLU":{"url":"https://vmlu.ai/leaderboard","metric":"Average (%)","theoretical_min":0.0,"theoretical_max":100.0},"VMLU - Humanities":{"url":"https://vmlu.ai/leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"VMLU - Other":{"url":"https://vmlu.ai/leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"VMLU - STEM":{"url":"https://vmlu.ai/leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"VMLU - Social Science":{"url":"https://vmlu.ai/leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"VNBench":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"Accuracy","theoretical_min":0.0,"theoretical_max":100.0},"VNTL Leaderboard":{"url":"https://huggingface.co/datasets/lmg-anon/vntl-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"VPCT":{"url":"https://epoch.ai/benchmarks/vpct","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"VRBench":{"url":"https://benchmarklist.com/benchmarks/vrbench/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"VSI-Super-Wild":{"url":"https://vsi-super-wild.github.io/","metric":"Overall Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"VTB":{"url":"https://benchmarklist.com/benchmarks/vtb/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Vals AI (Vals Index)":{"url":"https://www.vals.ai/benchmarks","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI AIME":{"url":"https://www.vals.ai/benchmarks/aime","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI CaseLaw v2":{"url":"https://www.vals.ai/benchmarks/case_law_v2","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"sources":[{"name":"Vals AI CaseLaw v2","url":"https://www.vals.ai/benchmarks/case_law_v2","metric":"Accuracy (%)"},{"name":"CaseLaw v2","url":"https://benchmarklist.com/benchmarks/vals_case_law_v2/","metric":"Score (self-reported)"}]},"Vals AI Code Migration":{"url":"https://www.vals.ai/benchmarks/code-migration","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI CorpFin v2":{"url":"https://www.vals.ai/benchmarks/corp_fin_v2","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI CyberBench":{"url":"https://www.vals.ai/benchmarks/cyber","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI Excel Modeling":{"url":"https://www.vals.ai/benchmarks/emb","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI Finance Agent":{"url":"https://www.vals.ai/benchmarks/finance_agent","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI Finance Agent v2":{"url":"https://www.vals.ai/benchmarks/fabv2","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI GPQA":{"url":"https://www.vals.ai/benchmarks/gpqa","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI Harvey Legal Agent Bench":{"url":"https://www.vals.ai/benchmarks/hlab","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI IOI":{"url":"https://www.vals.ai/benchmarks/ioi","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI Legal Research Bench":{"url":"https://www.vals.ai/benchmarks/legal_research","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI LegalBench":{"url":"https://www.vals.ai/benchmarks/legal_bench","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI LiveCodeBench":{"url":"https://www.vals.ai/benchmarks/lcb","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI MATH 500":{"url":"https://www.vals.ai/benchmarks/math500","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"sources":[{"name":"Vals AI MATH 500","url":"https://www.vals.ai/benchmarks/math500","metric":"Accuracy (%)"},{"name":"MATH 500","url":"https://benchmarklist.com/benchmarks/vals_math500/","metric":"Score (self-reported)"}]},"Vals AI MGSM":{"url":"https://www.vals.ai/benchmarks/mgsm","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"sources":[{"name":"Vals AI MGSM","url":"https://www.vals.ai/benchmarks/mgsm","metric":"Accuracy (%)"},{"name":"MGSM","url":"https://benchmarklist.com/benchmarks/vals_mgsm/","metric":"Score (self-reported)"}]},"Vals AI MMLU-Pro":{"url":"https://www.vals.ai/benchmarks/mmlu_pro","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI MMMU":{"url":"https://www.vals.ai/benchmarks/mmmu","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI MedCode":{"url":"https://www.vals.ai/benchmarks/medcode","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI MedQA":{"url":"https://www.vals.ai/benchmarks/medqa","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI MedScribe":{"url":"https://www.vals.ai/benchmarks/medscribe","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI MortgageTax":{"url":"https://www.vals.ai/benchmarks/mortgage_tax","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI Multimodal Index":{"url":"https://www.vals.ai/benchmarks/vals_multimodal_index","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI Poker Agent":{"url":"https://www.vals.ai/benchmarks/poker_agent","metric":"TrueSkill Rating","theoretical_min":null,"theoretical_max":null},"Vals AI ProgramBench":{"url":"https://www.vals.ai/benchmarks/programbench","metric":"Raw Pass Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI ProofBench":{"url":"https://www.vals.ai/benchmarks/proof_bench","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"sources":[{"name":"Vals AI ProofBench","url":"https://www.vals.ai/benchmarks/proof_bench","metric":"Accuracy (%)"},{"name":"ProofBench","url":"https://benchmarklist.com/benchmarks/vals_proof_bench/","metric":"Score (self-reported)"},{"name":"Epoch AI - Proofbench","url":"https://epoch.ai/benchmarks/proofbench","metric":"Score"}]},"Vals AI Public Benefits Bench":{"url":"https://www.vals.ai/benchmarks/public-benefits-bench","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI SAGE":{"url":"https://www.vals.ai/benchmarks/sage","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI SWE-bench Verified":{"url":"https://www.vals.ai/benchmarks/swebench","metric":"Resolved (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI SkillsBench":{"url":"https://www.vals.ai/benchmarks/skillsbench","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI TaxEval v2":{"url":"https://www.vals.ai/benchmarks/tax_eval_v2","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI Terminal-Bench 2.0":{"url":"https://www.vals.ai/benchmarks/terminal-bench-2","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI Terminal-Bench 2.1":{"url":"https://www.vals.ai/benchmarks/terminal-bench-2-1","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals AI Vibe Code Bench":{"url":"https://www.vals.ai/benchmarks/vibe-code","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vals Index":{"url":"https://benchmarklist.com/benchmarks/vals_vals_index/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Vals Multimodal Index":{"url":"https://benchmarklist.com/benchmarks/vals_vals_multimodal_index/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Vectara Hallucination Leaderboard":{"url":"https://github.com/vectara/hallucination-leaderboard","metric":"Factual Consistency Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vector Eval - ARC-C":{"url":"https://huggingface.co/spaces/vector-institute/eval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vector Eval - ARC-E":{"url":"https://huggingface.co/spaces/vector-institute/eval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vector Eval - AgentHarm":{"url":"https://huggingface.co/spaces/vector-institute/eval-leaderboard","metric":"Average Harm Score (%) (lower is better)","theoretical_min":0.0,"theoretical_max":100.0},"Vector Eval - AgentHarm-Benign":{"url":"https://huggingface.co/spaces/vector-institute/eval-leaderboard","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vector Eval - DROP":{"url":"https://huggingface.co/spaces/vector-institute/eval-leaderboard","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vector Eval - GAIA":{"url":"https://huggingface.co/spaces/vector-institute/eval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vector Eval - GPQA-D":{"url":"https://huggingface.co/spaces/vector-institute/eval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vector Eval - GSM8K":{"url":"https://huggingface.co/spaces/vector-institute/eval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vector Eval - HellaSwag":{"url":"https://huggingface.co/spaces/vector-institute/eval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vector Eval - HumanEval":{"url":"https://huggingface.co/spaces/vector-institute/eval-leaderboard","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vector Eval - IFEval":{"url":"https://huggingface.co/spaces/vector-institute/eval-leaderboard","metric":"Final Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vector Eval - In-House-CTF":{"url":"https://huggingface.co/spaces/vector-institute/eval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vector Eval - InterCode-CTF":{"url":"https://huggingface.co/spaces/vector-institute/eval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vector Eval - MATH":{"url":"https://huggingface.co/spaces/vector-institute/eval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vector Eval - MMLU":{"url":"https://huggingface.co/spaces/vector-institute/eval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vector Eval - MMLU-Pro":{"url":"https://huggingface.co/spaces/vector-institute/eval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vector Eval - MMMU-MC":{"url":"https://huggingface.co/spaces/vector-institute/eval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vector Eval - MMMU-OE":{"url":"https://huggingface.co/spaces/vector-institute/eval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vector Eval - SWE-Bench":{"url":"https://huggingface.co/spaces/vector-institute/eval-leaderboard","metric":"Mean Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vector Eval - WinoGrande":{"url":"https://huggingface.co/spaces/vector-institute/eval-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vellum - AIME 2025":{"url":"https://www.vellum.ai/llm-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vellum - ARC-AGI-2":{"url":"https://www.vellum.ai/llm-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vellum - GPQA":{"url":"https://www.vellum.ai/llm-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":25},"Vellum - HumanEval":{"url":"https://www.vellum.ai/llm-leaderboard","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vellum - Humanity's Last Exam":{"url":"https://www.vellum.ai/llm-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vellum - LiveCodeBench":{"url":"https://www.vellum.ai/llm-leaderboard","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vellum - MATH":{"url":"https://www.vellum.ai/llm-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vellum - MMMLU":{"url":"https://www.vellum.ai/llm-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":25},"Vending-Bench":{"url":"https://andonlabs.com/evals/vending-bench","metric":"Net Worth ($)","theoretical_min":null,"theoretical_max":null},"Vending-Bench 2":{"url":"https://andonlabs.com/evals/vending-bench-2","metric":"Money Balance ($)","theoretical_min":null,"theoretical_max":null},"VerdictBench":{"url":"https://benchmarklist.com/benchmarks/verdictbench/","metric":"Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"VeriContest":{"url":"https://benchmarklist.com/benchmarks/vericontest_a_competitive_programming_benchmark_for_verifiable_code_generation/","metric":"End-to-End (end2end) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"VeriTrip":{"url":"https://benchmarklist.com/benchmarks/veritrip_a_verifiable_benchmark_for_travel_planning_agents_over_unstructured_web_corpora/","metric":"DR (Simple) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Verus-SpecBench":{"url":"https://benchmarklist.com/benchmarks/verus_specbench/","metric":"Pass@1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Video-Holmes":{"url":"https://benchmarklist.com/benchmarks/video_holmes/","metric":"Overall (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Video-MME-v2":{"url":"https://video-mme-v2.netlify.app/","metric":"Avg Accuracy w/o sub (%)","theoretical_min":0.0,"theoretical_max":100.0},"Video-MMMU":{"url":"https://videommmu.github.io/","metric":"Overall Score","theoretical_min":0.0,"theoretical_max":100.0},"VideoGameBench":{"url":"https://vgbench.com/","metric":"Completion Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"VideoMME w sub.":{"url":"https://benchmarklist.com/benchmarks/videomme_with_sub/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"VideoMME w/o sub.":{"url":"https://benchmarklist.com/benchmarks/videomme_no_sub/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"VinCLAT (Catalan)":{"url":"https://huggingface.co/spaces/projecte-aina/vinclat-leaderboard","metric":"Average Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vinoground (Group Score)":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"Group Score","theoretical_min":0.0,"theoretical_max":100.0},"Vinoground (Text Score)":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"Text Score","theoretical_min":0.0,"theoretical_max":100.0},"Vinoground (Video Score)":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"Video Score","theoretical_min":0.0,"theoretical_max":100.0},"Virology Benchmark":{"url":"https://www.virologytest.ai/","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Vision2Code":{"url":"https://benchmarklist.com/benchmarks/vision2code_a_multi_domain_benchmark_for_evaluating_image_to_code_generation/","metric":"Avg. (self-reported)","theoretical_min":0.0,"theoretical_max":null},"VisuLogic":{"url":"https://visulogic-benchmark.github.io/VisuLogic/","metric":"Overall Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Visual Acuity Test (4K)":{"url":"https://jonathanc.net/visual-acuity/","metric":"Mean Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"Visual Aesthetic Benchmark (VAB)":{"url":"https://benchmarklist.com/benchmarks/visual_aesthetic_benchmark_vab/","metric":"Overall Top-1 ap@1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Visual-Language Understanding":{"url":"https://benchmarklist.com/benchmarks/visual_language_understanding/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"VisualNeedle":{"url":"https://benchmarklist.com/benchmarks/visualneedle/","metric":"w/ Tools Acc. (%) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"VisualPuzzles":{"url":"https://neulab.github.io/VisualPuzzles/","metric":"Overall Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"VisualWebArena":{"url":"https://jykoh.com/vwa","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"VitaBench":{"url":"https://vitabench.github.io/","metric":"Cross-Scenario Avg@4 (%)","theoretical_min":0.0,"theoretical_max":100.0},"VitaBench 2.0":{"url":"https://benchmarklist.com/benchmarks/vitabench_2_0/","metric":"Avg@4 Full Context (self-reported)","theoretical_min":0.0,"theoretical_max":null},"VoxelBench":{"url":"https://voxelbench.ai/leaderboard","metric":"Rating","theoretical_min":null,"theoretical_max":null},"WDCD":{"url":"https://huggingface.co/spaces/winzheng-Lab/wdcd-leaderboard","metric":"DCD Score","theoretical_min":0.0,"theoretical_max":100.0},"WDCD R1 Understanding":{"url":"https://huggingface.co/spaces/winzheng-Lab/wdcd-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"WDCD R2 In-Document Resistance":{"url":"https://huggingface.co/spaces/winzheng-Lab/wdcd-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"WDCD R3 Pressure Integrity":{"url":"https://huggingface.co/spaces/winzheng-Lab/wdcd-leaderboard","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":200.0},"WMT25 English-Bhojpuri ESA":{"url":"https://github.com/wmt-conference/wmt25-general-mt","metric":"ESA Human Score","theoretical_min":0.0,"theoretical_max":100.0},"WMT25 English-Bhojpuri chrF++":{"url":"https://github.com/wmt-conference/wmt25-general-mt","metric":"chrF++","theoretical_min":0.0,"theoretical_max":100.0},"WMT25 English-Maasai ESA":{"url":"https://github.com/wmt-conference/wmt25-general-mt","metric":"ESA Human Score","theoretical_min":0.0,"theoretical_max":100.0},"WMT25 English-Maasai chrF++":{"url":"https://github.com/wmt-conference/wmt25-general-mt","metric":"chrF++","theoretical_min":0.0,"theoretical_max":100.0},"WMW (World Models in Words)":{"url":"https://benchmarklist.com/benchmarks/wmw_world_models_in_words/","metric":"Trace–ans. consistency (self-reported)","theoretical_min":0.0,"theoretical_max":null},"We-Math":{"url":"https://we-math.github.io/","metric":"Avg Strict (%)","theoretical_min":0.0,"theoretical_max":100.0},"WeaveBench":{"url":"https://weavebench.github.io/","metric":"PassRate (%)","theoretical_min":0.0,"theoretical_max":100.0},"Web-Bench":{"url":"https://huggingface.co/spaces/bytedance-research/Web-Bench-Leaderboard","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"WebApp1K":{"url":"https://huggingface.co/spaces/onekq-ai/WebApp1K-models-leaderboard","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"WebApp1K Duo":{"url":"https://huggingface.co/spaces/onekq-ai/WebApp1K-models-leaderboard","metric":"Pass@1 (%)","theoretical_min":0.0,"theoretical_max":100.0},"WebApp1K-React":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"pass@1","theoretical_min":0.0,"theoretical_max":1.0},"WebApp1k-Duo-React":{"url":"https://huggingface.co/datasets/felixleungsc/paperswithcode-data-evaluation-tables","metric":"pass@1","theoretical_min":0.0,"theoretical_max":1.0},"WebArena":{"url":"https://docs.google.com/spreadsheets/d/1M801lEpBbKSNwP-vDBkC_pF7LdyGU1f_ufZb_NWNBZQ","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"WebArena Average@5 (Opus 4.6 System Card)":{"url":"https://www-cdn.anthropic.com/0dd865075ad3132672ee0ab40b05a53f14cf5288.pdf","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"WebCoderBench - Functional Alignment":{"url":"https://huggingface.co/spaces/WebCoderBench/WebCoderBench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"WebCoderBench - Functionality Correctness":{"url":"https://huggingface.co/spaces/WebCoderBench/WebCoderBench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"WebCoderBench - Performance":{"url":"https://huggingface.co/spaces/WebCoderBench/WebCoderBench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"WebCoderBench - Visual Experience":{"url":"https://huggingface.co/spaces/WebCoderBench/WebCoderBench","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"WebDev Arena":{"url":"https://arena.ai/leaderboard/code","metric":"Arena Score","theoretical_min":null,"theoretical_max":null},"WebVoyager":{"url":"https://leaderboard.steel.dev/leaderboards/webvoyager/","metric":"Success Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"WeirdML":{"url":"https://htihle.github.io/","metric":"Average Score","theoretical_min":0.0,"theoretical_max":100.0},"Werewolf Benchmark":{"url":"https://werewolf.foaster.ai/","metric":"ELO","theoretical_min":null,"theoretical_max":null},"Werewolf Benchmark - Win Rate":{"url":"https://werewolf.foaster.ai/","metric":"Win Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"When Context Flips, Safety Breaks":{"url":"https://benchmarklist.com/benchmarks/when_context_flips_safety_breaks_diagnosing_brittle_safety_in_aligned_language_models/","metric":"PacifAIst BSR (self-reported)","theoretical_min":0.0,"theoretical_max":null},"When Lower Privileges Suffice":{"url":"https://benchmarklist.com/benchmarks/when_lower_privileges_suffice_investigating_over_privileged_tool_selection_in_llm_agents/","metric":"Business (self-reported)","theoretical_min":0.0,"theoretical_max":null},"When Multiple Scripts Matter":{"url":"https://benchmarklist.com/benchmarks/when_multiple_scripts_matter_evaluating_asr_in_clinical_settings/","metric":"CER (original, original, original) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"When Reviews Disagree":{"url":"https://benchmarklist.com/benchmarks/when_reviews_disagree_fine_grained_contradiction_analysis_in_scientific_peer_reviews/","metric":"ρ (self-reported)","theoretical_min":0.0,"theoretical_max":null},"When Safe Skills Collide":{"url":"https://benchmarklist.com/benchmarks/when_safe_skills_collide_measuring_compositional_risk_in_agent_skill_ecosystems/","metric":"Full chain (self-reported)","theoretical_min":0.0,"theoretical_max":null},"When Seekers Are Hard to Help":{"url":"https://benchmarklist.com/benchmarks/when_seekers_are_hard_to_help_evaluating_emotional_support_dialogue_systems_in_worst_case_interactions/","metric":"DEU (Deep Emotional Understanding) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"When Simulation Lies":{"url":"https://benchmarklist.com/benchmarks/when_simulation_lies_a_sim_to_real_benchmark_and_domain_randomized_rl_recipe_for_tool_use_agents/","metric":"Pert Acc (self-reported)","theoretical_min":0.0,"theoretical_max":null},"When Symptoms Are Not Enough":{"url":"https://benchmarklist.com/benchmarks/when_symptoms_are_not_enough_evidence_weighting_patterns_in_large_language_model_psychiatric_screening/","metric":"Matthews correlation coefficient (self-reported)","theoretical_min":0.0,"theoretical_max":null},"When the Manual Lies":{"url":"https://benchmarklist.com/benchmarks/when_the_manual_lies_a_realistic_benchmark_to_evaluate_mcp_poisoning_attacks_for_llm_agents/","metric":"Attack Success Rate (ASR) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"When2Speak":{"url":"https://benchmarklist.com/benchmarks/when2speak_a_dataset_for_temporal_participation_and_turn_taking_in_multi_party_conversations_for_large_language_models/","metric":"0-shot Macro F1 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"WikiVQABench":{"url":"https://benchmarklist.com/benchmarks/wikivqabench_a_knowledge_grounded_visual_question_answering_benchmark_from_wikipedia_and_wikidata/","metric":"Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"WildAgtEval":{"url":"https://benchmarklist.com/benchmarks/wildagt_eval/","metric":"Complexity-Injected API Call Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"WildBench":{"url":"https://huggingface.co/spaces/allenai/WildBench","metric":"WB Score Task-Macro","theoretical_min":null,"theoretical_max":null},"WildClawBench":{"url":"https://benchmarklist.com/benchmarks/wildclawbench/","metric":"Overall Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"WildRoadBench":{"url":"https://benchmarklist.com/benchmarks/wildroadbench/","metric":"AP50 (self-reported)","theoretical_min":0.0,"theoretical_max":null},"WildVision-Bench":{"url":"https://huggingface.co/spaces/opencompass/openlmm_subjective_leaderboard","metric":"Overall Reward","theoretical_min":-100.0,"theoretical_max":100.0},"WinoGrande":{"url":"https://epoch.ai/benchmarks/eci","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":50},"WolfBench":{"url":"https://wolfbench.ai/","metric":"Average Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"Wolfram LLM Benchmarking Project":{"url":"https://www.wolfram.com/llm-benchmarking-project/","metric":"Correct Functionality (%)","theoretical_min":0.0,"theoretical_max":100.0},"Wordle Arena":{"url":"https://drchangliu.github.io/WordleArenas/","metric":"Win Rate (%)","theoretical_min":0.0,"theoretical_max":100.0},"WorldCupBench":{"url":"https://github.com/mverab/WorldCupBench","metric":"Quiniela Points","theoretical_min":0.0,"theoretical_max":null},"WorldCupBench - Brier Skill":{"url":"https://github.com/mverab/WorldCupBench","metric":"100 - Brier Total","theoretical_min":null,"theoretical_max":null},"WritingBench":{"url":"https://benchmarklist.com/benchmarks/writingbench/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"X-Risks Leaderboard":{"url":"https://huggingface.co/spaces/inceptionai/X-Risks-Leaderboard","metric":"3C3H Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"XDomainBench":{"url":"https://benchmarklist.com/benchmarks/xdomainbench/","metric":"R (k=1, Deterministic) (self-reported)","theoretical_min":0.0,"theoretical_max":null},"XL-SafetyBench":{"url":"https://benchmarklist.com/benchmarks/xl_safetybench_a_country_grounded_cross_cultural_benchmark_for_llm_safety_and_cultural_sensitivity/","metric":"Overall ASR (self-reported)","theoretical_min":0.0,"theoretical_max":null},"Xent Games":{"url":"https://benchmarklist.com/benchmarks/xent_games/","metric":"Overall Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"YC-Bench":{"url":"https://collinear-ai.github.io/yc-bench/","metric":"Net Worth ($K)","theoretical_min":0.0,"theoretical_max":null},"YKS 2025 LLM Leaderboard":{"url":"https://huggingface.co/spaces/atahanuz/YKS2025_LLM_Leaderboard","metric":"Total Score (out of 200)","theoretical_min":0.0,"theoretical_max":200.0},"YapBench":{"url":"https://huggingface.co/spaces/tabularisai/YapBench","metric":"YapIndex (lower is better)","theoretical_min":null,"theoretical_max":null},"You're Absolutely Right!":{"url":"https://typebulb.com/u/lab/you-re-absolutely-right/full","metric":"Average anti-sycophancy score (1-5)","theoretical_min":1.0,"theoretical_max":5.0},"ZEROBench-Sub":{"url":"https://benchmarklist.com/benchmarks/zerobench_sub/","metric":"Score (self-reported)","theoretical_min":0.0,"theoretical_max":null},"ZebraLogic":{"url":"https://huggingface.co/spaces/allenai/ZebraLogic","metric":"Puzzle Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0},"ZeroBench":{"url":"https://zerobench.github.io/","metric":"Score (%)","theoretical_min":0.0,"theoretical_max":100.0},"ZeroEval GPQA Diamond":{"url":"https://www.vals.ai/benchmarks/gpqa","metric":"GPQA Diamond Score","theoretical_min":0.0,"theoretical_max":100.0,"random_baseline":25},"ZeroEval MATH-500":{"url":"https://www.vals.ai/benchmarks/math500","metric":"MATH-500 Score","theoretical_min":0.0,"theoretical_max":100.0},"boogiebench":{"url":"https://www.boogiebench.com/leaderboard","metric":"Elo Rating","theoretical_min":null,"theoretical_max":null},"gg-bench":{"url":"https://github.com/vivek3141/gg-bench","metric":"Games Won","theoretical_min":0.0,"theoretical_max":100.0},"gwBenchmarks":{"url":"https://benchmarklist.com/benchmarks/gwbenchmarks/","metric":"Waveform (self-reported)","theoretical_min":0.0,"theoretical_max":null},"scBench":{"url":"https://benchmarklist.com/benchmarks/scbench/","metric":"Accuracy (self-reported)","theoretical_min":0.0,"theoretical_max":null},"μ-MATH":{"url":"https://huggingface.co/spaces/toloka/u-math-leaderboard","metric":"Accuracy (%)","theoretical_min":0.0,"theoretical_max":100.0}},"models":[{"model":"Gemini 2.5 Pro","provider":"Google","country":"US","openness":"api","release_date":"2025-03-25","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":1794,"benchmark_group_count":1331,"benchmark_coverage_score":719.0,"source_family_count":150,"source_group_family_count":150,"source_effective_breadth":6.69,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.2977,"source_primary_group_share":0.3418,"capability_breadth":3.9,"capability_primary":"vision_language","capability_primary_share":0.4223,"capability_primary_group_count":612,"capability_non_primary_group_count":719},{"model":"GPT-4o","provider":"OpenAI","country":"US","openness":"api","release_date":"2024-05-13","param_count":200.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":1474,"benchmark_group_count":1220,"benchmark_coverage_score":649.0,"source_family_count":153,"source_group_family_count":153,"source_effective_breadth":5.43,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.3768,"source_primary_group_share":0.382,"capability_breadth":3.84,"capability_primary":"vision_language","capability_primary_share":0.453,"capability_primary_group_count":571,"capability_non_primary_group_count":649},{"model":"Gemma 3 27B (IT)","provider":"Google","country":"US","openness":"open","release_date":"2025-03-12","param_count":27.0,"active_param_count":null,"training_compute_flop":2.268e+24,"num_benchmarks":1450,"benchmark_group_count":1009,"benchmark_coverage_score":558.0,"source_family_count":54,"source_group_family_count":54,"source_effective_breadth":3.96,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.3648,"source_primary_group_share":0.4509,"capability_breadth":3.51,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3796,"capability_primary_group_count":451,"capability_non_primary_group_count":558},{"model":"Gemma 3 12B (IT)","provider":"Google","country":"US","openness":"open","release_date":"2025-03-12","param_count":12.0,"active_param_count":null,"training_compute_flop":8.64e+23,"num_benchmarks":1154,"benchmark_group_count":770,"benchmark_coverage_score":320.0,"source_family_count":36,"source_group_family_count":36,"source_effective_breadth":2.61,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.4767,"source_primary_group_share":0.5909,"capability_breadth":3.35,"capability_primary":"vision_language","capability_primary_share":0.4584,"capability_primary_group_count":450,"capability_non_primary_group_count":320},{"model":"Gemma 3 4B (IT)","provider":"Google","country":"US","openness":"open","release_date":"2025-03-12","param_count":4.0,"active_param_count":null,"training_compute_flop":9.6e+22,"num_benchmarks":1154,"benchmark_group_count":773,"benchmark_coverage_score":324.0,"source_family_count":34,"source_group_family_count":34,"source_effective_breadth":2.63,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.4767,"source_primary_group_share":0.5886,"capability_breadth":3.34,"capability_primary":"vision_language","capability_primary_share":0.4575,"capability_primary_group_count":449,"capability_non_primary_group_count":324},{"model":"Llama 3.3 70B Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2024-12-06","param_count":70.0,"active_param_count":null,"training_compute_flop":6.8649768e+24,"num_benchmarks":1110,"benchmark_group_count":737,"benchmark_coverage_score":356.0,"source_family_count":78,"source_group_family_count":78,"source_effective_breadth":10.35,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.2094,"source_primary_group_share":0.228,"capability_breadth":3.02,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5335,"capability_primary_group_count":381,"capability_non_primary_group_count":356},{"model":"Llama 3.1 8B Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-23","param_count":8.0,"active_param_count":null,"training_compute_flop":1.224e+24,"num_benchmarks":1071,"benchmark_group_count":637,"benchmark_coverage_score":408.0,"source_family_count":96,"source_group_family_count":96,"source_effective_breadth":18.38,"source_primary":"euroeval.com","source_primary_share":0.201,"source_primary_group_share":0.1444,"capability_breadth":4.06,"capability_primary":"knowledge_multilingual","capability_primary_share":0.402,"capability_primary_group_count":229,"capability_non_primary_group_count":408},{"model":"GPT-4.1 Mini","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-04-14","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":1062,"benchmark_group_count":744,"benchmark_coverage_score":387.0,"source_family_count":73,"source_group_family_count":73,"source_effective_breadth":5.45,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.3372,"source_primary_group_share":0.3871,"capability_breadth":3.98,"capability_primary":"vision_language","capability_primary_share":0.4425,"capability_primary_group_count":357,"capability_non_primary_group_count":387},{"model":"InternVL3-78B","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2025-04-11","param_count":78.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":1059,"benchmark_group_count":846,"benchmark_coverage_score":45.0,"source_family_count":21,"source_group_family_count":21,"source_effective_breadth":2.45,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.5044,"source_primary_group_share":0.5378,"capability_breadth":1.18,"capability_primary":"vision_language","capability_primary_share":0.9206,"capability_primary_group_count":801,"capability_non_primary_group_count":45},{"model":"InternVL2-8B","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2024-07-15","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":1056,"benchmark_group_count":858,"benchmark_coverage_score":40.0,"source_family_count":14,"source_group_family_count":14,"source_effective_breadth":2.41,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.5131,"source_primary_group_share":0.5431,"capability_breadth":1.16,"capability_primary":"vision_language","capability_primary_share":0.9292,"capability_primary_group_count":818,"capability_non_primary_group_count":40},{"model":"InternVL3-8B","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2025-04-11","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":1048,"benchmark_group_count":842,"benchmark_coverage_score":33.0,"source_family_count":17,"source_group_family_count":17,"source_effective_breadth":2.42,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.5098,"source_primary_group_share":0.5404,"capability_breadth":1.14,"capability_primary":"vision_language","capability_primary_share":0.9354,"capability_primary_group_count":809,"capability_non_primary_group_count":33},{"model":"Qwen 2 VL 72B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-19","param_count":72.0,"active_param_count":null,"training_compute_flop":6.048e+23,"num_benchmarks":1046,"benchmark_group_count":848,"benchmark_coverage_score":42.0,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":2.35,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.5181,"source_primary_group_share":0.5495,"capability_breadth":1.16,"capability_primary":"vision_language","capability_primary_share":0.9265,"capability_primary_group_count":806,"capability_non_primary_group_count":42},{"model":"Qwen 2 VL 7B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-08-29","param_count":7.0,"active_param_count":null,"training_compute_flop":6.72e+22,"num_benchmarks":1046,"benchmark_group_count":848,"benchmark_coverage_score":40.0,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":2.35,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.5181,"source_primary_group_share":0.5495,"capability_breadth":1.16,"capability_primary":"vision_language","capability_primary_share":0.9285,"capability_primary_group_count":808,"capability_non_primary_group_count":40},{"model":"InternVL3-38B","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2025-04-11","param_count":38.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":1034,"benchmark_group_count":829,"benchmark_coverage_score":33.0,"source_family_count":16,"source_group_family_count":16,"source_effective_breadth":2.36,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.5164,"source_primary_group_share":0.5489,"capability_breadth":1.14,"capability_primary":"vision_language","capability_primary_share":0.9346,"capability_primary_group_count":796,"capability_non_primary_group_count":33},{"model":"Qwen 2 VL 2B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-08-29","param_count":2.0,"active_param_count":null,"training_compute_flop":1.68e+22,"num_benchmarks":1028,"benchmark_group_count":831,"benchmark_coverage_score":40.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.26,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.5269,"source_primary_group_share":0.5608,"capability_breadth":1.16,"capability_primary":"vision_language","capability_primary_share":0.9273,"capability_primary_group_count":791,"capability_non_primary_group_count":40},{"model":"Claude Sonnet 4.6","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-02-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":1026,"benchmark_group_count":756,"benchmark_coverage_score":464.0,"source_family_count":101,"source_group_family_count":99,"source_effective_breadth":8.8,"source_primary":"euroeval.com","source_primary_share":0.2608,"source_primary_group_share":0.2209,"capability_breadth":4.07,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4504,"capability_primary_group_count":292,"capability_non_primary_group_count":464},{"model":"InternVL3-14B","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2025-04-11","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":1018,"benchmark_group_count":812,"benchmark_coverage_score":31.0,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":2.26,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.5252,"source_primary_group_share":0.5603,"capability_breadth":1.14,"capability_primary":"vision_language","capability_primary_share":0.9355,"capability_primary_group_count":781,"capability_non_primary_group_count":31},{"model":"Claude 3.5 Sonnet (20241022)","provider":"Anthropic","country":"US","openness":"api","release_date":"2024-10-22","param_count":null,"active_param_count":null,"training_compute_flop":2.7e+25,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 3 family retired or superseded by newer Claude models.","num_benchmarks":1011,"benchmark_group_count":784,"benchmark_coverage_score":186.0,"source_family_count":48,"source_group_family_count":48,"source_effective_breadth":2.66,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.5387,"source_primary_group_share":0.5944,"capability_breadth":1.74,"capability_primary":"vision_language","capability_primary_share":0.7515,"capability_primary_group_count":598,"capability_non_primary_group_count":186},{"model":"Ovis2-8B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-02-21","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":1008,"benchmark_group_count":805,"benchmark_coverage_score":31.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":2.22,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.5289,"source_primary_group_share":0.5652,"capability_breadth":1.14,"capability_primary":"vision_language","capability_primary_share":0.935,"capability_primary_group_count":774,"capability_non_primary_group_count":31},{"model":"GPT-4.1","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-04-14","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":1000,"benchmark_group_count":726,"benchmark_coverage_score":427.0,"source_family_count":115,"source_group_family_count":115,"source_effective_breadth":9.79,"source_primary":"euroeval.com","source_primary_share":0.2664,"source_primary_group_share":0.2314,"capability_breadth":3.93,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4672,"capability_primary_group_count":299,"capability_non_primary_group_count":427},{"model":"Aquila-VL-2B","provider":"BAAI","country":"CN","openness":"open","release_date":"2024-10-30","param_count":2.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":1000,"benchmark_group_count":803,"benchmark_coverage_score":35.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.13,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.542,"source_primary_group_share":0.5803,"capability_breadth":1.15,"capability_primary":"vision_language","capability_primary_share":0.9303,"capability_primary_group_count":768,"capability_non_primary_group_count":35},{"model":"MiniCPM-V-2.6","provider":"OpenBMB","country":"CN","openness":"open","release_date":"2024-08-06","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":998,"benchmark_group_count":811,"benchmark_coverage_score":30.0,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":2.18,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.5415,"source_primary_group_share":0.5746,"capability_breadth":1.13,"capability_primary":"vision_language","capability_primary_share":0.9386,"capability_primary_group_count":781,"capability_non_primary_group_count":30},{"model":"GPT-4o Mini","provider":"OpenAI","country":"US","openness":"api","release_date":"2024-07-18","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":957,"benchmark_group_count":814,"benchmark_coverage_score":329.0,"source_family_count":87,"source_group_family_count":87,"source_effective_breadth":2.9,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.5731,"source_primary_group_share":0.5725,"capability_breadth":2.75,"capability_primary":"vision_language","capability_primary_share":0.5861,"capability_primary_group_count":485,"capability_non_primary_group_count":329},{"model":"Gemini 3.1 Pro (Preview)","provider":"Google","country":"US","openness":"api","release_date":"2026-02-19","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":951,"benchmark_group_count":779,"benchmark_coverage_score":584.0,"source_family_count":159,"source_group_family_count":156,"source_effective_breadth":9.97,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.247,"source_primary_group_share":0.2157,"capability_breadth":6.46,"capability_primary":"knowledge_multilingual","capability_primary_share":0.295,"capability_primary_group_count":194,"capability_non_primary_group_count":584},{"model":"Gemini 1.5 Pro (002)","provider":"Google","country":"US","openness":"api","release_date":"2024-09-24","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 1.x family retired or superseded by newer Gemini models.","num_benchmarks":949,"benchmark_group_count":702,"benchmark_coverage_score":132.0,"source_family_count":25,"source_group_family_count":25,"source_effective_breadth":2.16,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.5694,"source_primary_group_share":0.6638,"capability_breadth":1.68,"capability_primary":"vision_language","capability_primary_share":0.7653,"capability_primary_group_count":570,"capability_non_primary_group_count":132},{"model":"GPT-4.1 Nano","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-04-14","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":946,"benchmark_group_count":637,"benchmark_coverage_score":285.0,"source_family_count":58,"source_group_family_count":58,"source_effective_breadth":4.2,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.3769,"source_primary_group_share":0.4521,"capability_breadth":3.37,"capability_primary":"vision_language","capability_primary_share":0.4893,"capability_primary_group_count":352,"capability_non_primary_group_count":285},{"model":"Pixtral-12B","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-09-17","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":946,"benchmark_group_count":802,"benchmark_coverage_score":49.0,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":2.12,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.5725,"source_primary_group_share":0.581,"capability_breadth":1.16,"capability_primary":"vision_language","capability_primary_share":0.9297,"capability_primary_group_count":753,"capability_non_primary_group_count":49},{"model":"GPT-5","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-08-07","param_count":null,"active_param_count":null,"training_compute_flop":6.6e+25,"num_benchmarks":918,"benchmark_group_count":671,"benchmark_coverage_score":395.0,"source_family_count":122,"source_group_family_count":122,"source_effective_breadth":9.33,"source_primary":"euroeval.com","source_primary_share":0.2608,"source_primary_group_share":0.2504,"capability_breadth":3.78,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4767,"capability_primary_group_count":275,"capability_non_primary_group_count":395},{"model":"GPT-5.2","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-12-11","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":864,"benchmark_group_count":618,"benchmark_coverage_score":335.0,"source_family_count":92,"source_group_family_count":92,"source_effective_breadth":7.5,"source_primary":"euroeval.com","source_primary_share":0.3081,"source_primary_group_share":0.2718,"capability_breadth":3.22,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5261,"capability_primary_group_count":283,"capability_non_primary_group_count":335},{"model":"Gemini 1.5 Flash (002)","provider":"Google","country":"US","openness":"api","release_date":"2024-09-24","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 1.x family retired or superseded by newer Gemini models.","num_benchmarks":840,"benchmark_group_count":654,"benchmark_coverage_score":112.0,"source_family_count":17,"source_group_family_count":17,"source_effective_breadth":1.89,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.642,"source_primary_group_share":0.7125,"capability_breadth":1.6,"capability_primary":"vision_language","capability_primary_share":0.784,"capability_primary_group_count":542,"capability_non_primary_group_count":112},{"model":"InternVL2.5-78B","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2024-12-05","param_count":78.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":826,"benchmark_group_count":636,"benchmark_coverage_score":37.0,"source_family_count":15,"source_group_family_count":15,"source_effective_breadth":1.78,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.6563,"source_primary_group_share":0.7327,"capability_breadth":1.2,"capability_primary":"vision_language","capability_primary_share":0.9132,"capability_primary_group_count":599,"capability_non_primary_group_count":37},{"model":"GPT-5 Nano","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-08-07","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":800,"benchmark_group_count":510,"benchmark_coverage_score":223.0,"source_family_count":53,"source_group_family_count":53,"source_effective_breadth":5.26,"source_primary":"euroeval.com","source_primary_share":0.3283,"source_primary_group_share":0.3294,"capability_breadth":2.76,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5606,"capability_primary_group_count":287,"capability_non_primary_group_count":223},{"model":"Llama 3.1 70B Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-23","param_count":70.0,"active_param_count":null,"training_compute_flop":7.929e+24,"num_benchmarks":780,"benchmark_group_count":597,"benchmark_coverage_score":289.0,"source_family_count":77,"source_group_family_count":77,"source_effective_breadth":9.06,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.2957,"source_primary_group_share":0.2814,"capability_breadth":2.97,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5525,"capability_primary_group_count":308,"capability_non_primary_group_count":289},{"model":"InternVL2.5-2B","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2024-12-06","param_count":2.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":780,"benchmark_group_count":591,"benchmark_coverage_score":35.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":1.55,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.6961,"source_primary_group_share":0.7885,"capability_breadth":1.2,"capability_primary":"vision_language","capability_primary_share":0.9105,"capability_primary_group_count":556,"capability_non_primary_group_count":35},{"model":"Claude Sonnet 4.5","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-09-29","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":777,"benchmark_group_count":664,"benchmark_coverage_score":474.0,"source_family_count":119,"source_group_family_count":119,"source_effective_breadth":10.39,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.3016,"source_primary_group_share":0.253,"capability_breadth":5.87,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3333,"capability_primary_group_count":190,"capability_non_primary_group_count":474},{"model":"GPT-5 Mini","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-08-07","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":758,"benchmark_group_count":531,"benchmark_coverage_score":244.0,"source_family_count":73,"source_group_family_count":73,"source_effective_breadth":6.02,"source_primary":"euroeval.com","source_primary_share":0.3471,"source_primary_group_share":0.3164,"capability_breadth":2.68,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5928,"capability_primary_group_count":287,"capability_non_primary_group_count":244},{"model":"GPT-5.5","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-04-23","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":746,"benchmark_group_count":630,"benchmark_coverage_score":446.0,"source_family_count":119,"source_group_family_count":117,"source_effective_breadth":9.48,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.3175,"source_primary_group_share":0.2667,"capability_breadth":5.94,"capability_primary":"knowledge_multilingual","capability_primary_share":0.338,"capability_primary_group_count":179,"capability_non_primary_group_count":446},{"model":"Qwen 3 30B A3B 2507 Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-07-01","param_count":30.0,"active_param_count":3.0,"training_compute_flop":6.48e+23,"num_benchmarks":729,"benchmark_group_count":394,"benchmark_coverage_score":97.0,"source_family_count":26,"source_group_family_count":26,"source_effective_breadth":3.49,"source_primary":"euroeval.com","source_primary_share":0.3571,"source_primary_group_share":0.4264,"capability_breadth":2.0,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6786,"capability_primary_group_count":297,"capability_non_primary_group_count":97},{"model":"Llama 3.1 8B","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-23","param_count":8.0,"active_param_count":null,"training_compute_flop":1.224e+24,"num_benchmarks":721,"benchmark_group_count":356,"benchmark_coverage_score":158.0,"source_family_count":40,"source_group_family_count":40,"source_effective_breadth":7.14,"source_primary":"euroeval.com","source_primary_share":0.3516,"source_primary_group_share":0.3006,"capability_breadth":2.78,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5297,"capability_primary_group_count":198,"capability_non_primary_group_count":158},{"model":"Claude 3.7 Sonnet","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-02-24","param_count":null,"active_param_count":null,"training_compute_flop":3.35e+25,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 3 family retired or superseded by newer Claude models.","num_benchmarks":718,"benchmark_group_count":557,"benchmark_coverage_score":201.0,"source_family_count":75,"source_group_family_count":75,"source_effective_breadth":3.43,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.5029,"source_primary_group_share":0.5135,"capability_breadth":2.23,"capability_primary":"vision_language","capability_primary_share":0.6595,"capability_primary_group_count":356,"capability_non_primary_group_count":201},{"model":"Gemini 2.5 Flash","provider":"Google","country":"US","openness":"api","release_date":"2025-06-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":701,"benchmark_group_count":582,"benchmark_coverage_score":391.0,"source_family_count":104,"source_group_family_count":104,"source_effective_breadth":7.97,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.3368,"source_primary_group_share":0.2887,"capability_breadth":5.28,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3737,"capability_primary_group_count":191,"capability_non_primary_group_count":391},{"model":"Claude Opus 4.7","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-04-16","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":697,"benchmark_group_count":589,"benchmark_coverage_score":400.0,"source_family_count":101,"source_group_family_count":99,"source_effective_breadth":7.97,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.3373,"source_primary_group_share":0.2835,"capability_breadth":5.26,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3701,"capability_primary_group_count":186,"capability_non_primary_group_count":400},{"model":"Qwen 2.5 72B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-19","param_count":72.0,"active_param_count":null,"training_compute_flop":7.8e+24,"num_benchmarks":695,"benchmark_group_count":411,"benchmark_coverage_score":246.0,"source_family_count":68,"source_group_family_count":68,"source_effective_breadth":11.08,"source_primary":"euroeval.com","source_primary_share":0.3023,"source_primary_group_share":0.2238,"capability_breadth":3.84,"capability_primary":"knowledge_multilingual","capability_primary_share":0.423,"capability_primary_group_count":165,"capability_non_primary_group_count":246},{"model":"GPT-5.4","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-03-06","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":694,"benchmark_group_count":582,"benchmark_coverage_score":407.0,"source_family_count":92,"source_group_family_count":91,"source_effective_breadth":6.12,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.3403,"source_primary_group_share":0.2869,"capability_breadth":5.6,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3523,"capability_primary_group_count":175,"capability_non_primary_group_count":407},{"model":"Claude 3.5 Sonnet (20240620)","provider":"Anthropic","country":"US","openness":"api","release_date":"2024-06-20","param_count":null,"active_param_count":null,"training_compute_flop":2.7e+25,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 3 family retired or superseded by newer Claude models.","num_benchmarks":687,"benchmark_group_count":583,"benchmark_coverage_score":133.0,"source_family_count":35,"source_group_family_count":35,"source_effective_breadth":1.56,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.7943,"source_primary_group_share":0.7993,"capability_breadth":1.74,"capability_primary":"vision_language","capability_primary_share":0.7523,"capability_primary_group_count":450,"capability_non_primary_group_count":133},{"model":"Ministral 3 8B","provider":"Mistral","country":"FR","openness":"open","release_date":"2025-12-02","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":649,"benchmark_group_count":399,"benchmark_coverage_score":119.0,"source_family_count":17,"source_group_family_count":17,"source_effective_breadth":3.38,"source_primary":"euroeval.com","source_primary_share":0.4037,"source_primary_group_share":0.4211,"capability_breadth":2.02,"capability_primary":"knowledge_multilingual","capability_primary_share":0.691,"capability_primary_group_count":280,"capability_non_primary_group_count":119},{"model":"Llama 3.2 11B Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2024-09-25","param_count":11.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":639,"benchmark_group_count":493,"benchmark_coverage_score":125.0,"source_family_count":18,"source_group_family_count":17,"source_effective_breadth":2.61,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.5589,"source_primary_group_share":0.5842,"capability_breadth":1.81,"capability_primary":"vision_language","capability_primary_share":0.7373,"capability_primary_group_count":368,"capability_non_primary_group_count":125},{"model":"Llama 4 Scout Base","provider":"Meta","country":"US","openness":"open","release_date":"2025-04-05","param_count":109.0,"active_param_count":17.0,"training_compute_flop":4.08e+24,"num_benchmarks":634,"benchmark_group_count":501,"benchmark_coverage_score":63.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.2,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.843,"source_primary_group_share":0.9082,"capability_breadth":1.56,"capability_primary":"vision_language","capability_primary_share":0.7896,"capability_primary_group_count":438,"capability_non_primary_group_count":63},{"model":"Phi-4","provider":"Microsoft","country":"US","openness":"open","release_date":"2024-12-12","param_count":14.0,"active_param_count":null,"training_compute_flop":9.32e+23,"num_benchmarks":625,"benchmark_group_count":387,"benchmark_coverage_score":135.0,"source_family_count":54,"source_group_family_count":54,"source_effective_breadth":4.51,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.3671,"source_primary_group_share":0.4341,"capability_breadth":2.41,"capability_primary":"knowledge_multilingual","capability_primary_share":0.599,"capability_primary_group_count":252,"capability_non_primary_group_count":135},{"model":"Gemma 3 27B","provider":"Google","country":"US","openness":"open","release_date":"2025-03-12","param_count":27.0,"active_param_count":null,"training_compute_flop":2.268e+24,"num_benchmarks":616,"benchmark_group_count":463,"benchmark_coverage_score":115.0,"source_family_count":38,"source_group_family_count":37,"source_effective_breadth":2.46,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.5867,"source_primary_group_share":0.622,"capability_breadth":1.72,"capability_primary":"vision_language","capability_primary_share":0.7533,"capability_primary_group_count":348,"capability_non_primary_group_count":115},{"model":"Llama 4 Maverick","provider":"Meta","country":"US","openness":"open","release_date":"2025-04-05","param_count":400.0,"active_param_count":17.0,"training_compute_flop":2.244e+24,"num_benchmarks":609,"benchmark_group_count":470,"benchmark_coverage_score":279.0,"source_family_count":84,"source_group_family_count":84,"source_effective_breadth":6.01,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.3826,"source_primary_group_share":0.3574,"capability_breadth":4.35,"capability_primary":"knowledge_multilingual","capability_primary_share":0.437,"capability_primary_group_count":190,"capability_non_primary_group_count":279},{"model":"Grok 2 (1212)","provider":"xAI","country":"US","openness":"api","release_date":"2024-12-12","param_count":270.0,"active_param_count":null,"training_compute_flop":2.96e+25,"num_benchmarks":605,"benchmark_group_count":432,"benchmark_coverage_score":96.0,"source_family_count":25,"source_group_family_count":25,"source_effective_breadth":2.14,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.5889,"source_primary_group_share":0.6644,"capability_breadth":1.82,"capability_primary":"vision_language","capability_primary_share":0.7265,"capability_primary_group_count":336,"capability_non_primary_group_count":96},{"model":"Gemini 1.5 Pro","provider":"Google","country":"US","openness":"api","release_date":"2024-02-15","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 1.x family retired or superseded by newer Gemini models.","num_benchmarks":599,"benchmark_group_count":495,"benchmark_coverage_score":141.0,"source_family_count":72,"source_group_family_count":72,"source_effective_breadth":2.81,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.6066,"source_primary_group_share":0.5879,"capability_breadth":1.89,"capability_primary":"vision_language","capability_primary_share":0.7192,"capability_primary_group_count":354,"capability_non_primary_group_count":141},{"model":"Mistral Small 3.2","provider":"Mistral","country":"FR","openness":"open","release_date":"2025-06-20","param_count":24.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":598,"benchmark_group_count":350,"benchmark_coverage_score":76.0,"source_family_count":24,"source_group_family_count":24,"source_effective_breadth":2.85,"source_primary":"euroeval.com","source_primary_share":0.4355,"source_primary_group_share":0.48,"capability_breadth":1.8,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7353,"capability_primary_group_count":274,"capability_non_primary_group_count":76},{"model":"Ministral 3 14B","provider":"Mistral","country":"FR","openness":"open","release_date":"2025-12-02","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":595,"benchmark_group_count":348,"benchmark_coverage_score":70.0,"source_family_count":20,"source_group_family_count":20,"source_effective_breadth":2.81,"source_primary":"euroeval.com","source_primary_share":0.4384,"source_primary_group_share":0.4828,"capability_breadth":1.75,"capability_primary":"knowledge_multilingual","capability_primary_share":0.747,"capability_primary_group_count":278,"capability_non_primary_group_count":70},{"model":"Claude 3.5 Sonnet","provider":"Anthropic","country":"US","openness":"api","release_date":"2024-06-20","param_count":null,"active_param_count":null,"training_compute_flop":2.7e+25,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 3 family retired or superseded by newer Claude models.","num_benchmarks":591,"benchmark_group_count":491,"benchmark_coverage_score":160.0,"source_family_count":79,"source_group_family_count":79,"source_effective_breadth":2.78,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.6059,"source_primary_group_share":0.5906,"capability_breadth":2.11,"capability_primary":"vision_language","capability_primary_share":0.6788,"capability_primary_group_count":331,"capability_non_primary_group_count":160},{"model":"Gemini 2.0 Flash (Preview)","provider":"Google","country":"US","openness":"api","release_date":"2024-12-11","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 2.0 family deprecated in favor of newer Gemini models.","num_benchmarks":587,"benchmark_group_count":501,"benchmark_coverage_score":45.0,"source_family_count":23,"source_group_family_count":23,"source_effective_breadth":1.21,"source_primary":"huggingface.co/spaces/tiger-lab/mega-bench","source_primary_share":0.914,"source_primary_group_share":0.9082,"capability_breadth":1.25,"capability_primary":"vision_language","capability_primary_share":0.893,"capability_primary_group_count":456,"capability_non_primary_group_count":45},{"model":"Llama 3.2 3B Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2024-09-25","param_count":3.0,"active_param_count":null,"training_compute_flop":1.733e+23,"num_benchmarks":587,"benchmark_group_count":346,"benchmark_coverage_score":177.0,"source_family_count":48,"source_group_family_count":48,"source_effective_breadth":7.88,"source_primary":"euroeval.com","source_primary_share":0.3414,"source_primary_group_share":0.2572,"capability_breadth":3.1,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5052,"capability_primary_group_count":169,"capability_non_primary_group_count":177},{"model":"Qwen 3 8B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":8.0,"active_param_count":null,"training_compute_flop":1.771e+24,"num_benchmarks":581,"benchmark_group_count":286,"benchmark_coverage_score":184.0,"source_family_count":45,"source_group_family_count":45,"source_effective_breadth":6.77,"source_primary":"euroeval.com","source_primary_share":0.2472,"source_primary_group_share":0.2657,"capability_breadth":3.78,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3755,"capability_primary_group_count":102,"capability_non_primary_group_count":184},{"model":"Gemma 3 12B","provider":"Google","country":"US","openness":"open","release_date":"2025-03-12","param_count":12.0,"active_param_count":null,"training_compute_flop":8.64e+23,"num_benchmarks":574,"benchmark_group_count":426,"benchmark_coverage_score":86.0,"source_family_count":22,"source_group_family_count":21,"source_effective_breadth":2.1,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.6275,"source_primary_group_share":0.6761,"capability_breadth":1.57,"capability_primary":"vision_language","capability_primary_share":0.7914,"capability_primary_group_count":340,"capability_non_primary_group_count":86},{"model":"Gemini 1.5 Flash","provider":"Google","country":"US","openness":"api","release_date":"2024-05-15","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 1.x family retired or superseded by newer Gemini models.","num_benchmarks":567,"benchmark_group_count":463,"benchmark_coverage_score":126.0,"source_family_count":49,"source_group_family_count":49,"source_effective_breadth":2.45,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.6459,"source_primary_group_share":0.6328,"capability_breadth":1.85,"capability_primary":"vision_language","capability_primary_share":0.7284,"capability_primary_group_count":337,"capability_non_primary_group_count":126},{"model":"Gemma 3 4B","provider":"Google","country":"US","openness":"open","release_date":"2025-03-12","param_count":4.0,"active_param_count":null,"training_compute_flop":9.6e+22,"num_benchmarks":567,"benchmark_group_count":419,"benchmark_coverage_score":82.0,"source_family_count":22,"source_group_family_count":21,"source_effective_breadth":2.04,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.6342,"source_primary_group_share":0.6874,"capability_breadth":1.56,"capability_primary":"vision_language","capability_primary_share":0.7946,"capability_primary_group_count":337,"capability_non_primary_group_count":82},{"model":"GPT-OSS-120B","provider":"OpenAI","country":"US","openness":"open","release_date":"2025-08-05","param_count":120.0,"active_param_count":5.13,"training_compute_flop":4.94e+24,"num_benchmarks":564,"benchmark_group_count":454,"benchmark_coverage_score":268.0,"source_family_count":100,"source_group_family_count":99,"source_effective_breadth":6.37,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.413,"source_primary_group_share":0.37,"capability_breadth":4.15,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4465,"capability_primary_group_count":185,"capability_non_primary_group_count":268},{"model":"Llama 3 70B Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2024-04-18","param_count":70.0,"active_param_count":null,"training_compute_flop":7.861e+24,"num_benchmarks":563,"benchmark_group_count":405,"benchmark_coverage_score":150.0,"source_family_count":52,"source_group_family_count":51,"source_effective_breadth":5.13,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.4071,"source_primary_group_share":0.4148,"capability_breadth":2.28,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6429,"capability_primary_group_count":255,"capability_non_primary_group_count":150},{"model":"Claude Sonnet 4","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-05-22","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 4.0 family deprecated in favor of newer Claude 4.x models.","num_benchmarks":557,"benchmark_group_count":453,"benchmark_coverage_score":278.0,"source_family_count":86,"source_group_family_count":86,"source_effective_breadth":5.77,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.4191,"source_primary_group_share":0.3709,"capability_breadth":4.4,"capability_primary":"knowledge_multilingual","capability_primary_share":0.432,"capability_primary_group_count":175,"capability_non_primary_group_count":278},{"model":"granite-4.0-h-micro","provider":"IBM","country":"US","openness":"open","release_date":"2025-10-01","param_count":3.0,"active_param_count":null,"training_compute_flop":3.15e+23,"num_benchmarks":551,"benchmark_group_count":351,"benchmark_coverage_score":80.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.69,"source_primary":"euroeval.com","source_primary_share":0.4589,"source_primary_group_share":0.4786,"capability_breadth":1.65,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7715,"capability_primary_group_count":271,"capability_non_primary_group_count":80},{"model":"Qwen 3.6 Plus","provider":"Alibaba","country":"CN","openness":"api","release_date":"2026-04-02","param_count":700.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":548,"benchmark_group_count":422,"benchmark_coverage_score":252.0,"source_family_count":44,"source_group_family_count":42,"source_effective_breadth":4.77,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.4222,"source_primary_group_share":0.3981,"capability_breadth":4.16,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4323,"capability_primary_group_count":169,"capability_non_primary_group_count":252},{"model":"Gemini 2.0 Flash","provider":"Google","country":"US","openness":"api","release_date":"2024-12-11","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 2.0 family deprecated in favor of newer Gemini models.","num_benchmarks":534,"benchmark_group_count":379,"benchmark_coverage_score":209.0,"source_family_count":70,"source_group_family_count":70,"source_effective_breadth":8.09,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.2967,"source_primary_group_share":0.2427,"capability_breadth":3.17,"capability_primary":"vision_language","capability_primary_share":0.5299,"capability_primary_group_count":170,"capability_non_primary_group_count":209},{"model":"Claude Haiku 4.5","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-10-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":529,"benchmark_group_count":437,"benchmark_coverage_score":252.0,"source_family_count":70,"source_group_family_count":70,"source_effective_breadth":5.08,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.4444,"source_primary_group_share":0.3844,"capability_breadth":3.84,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4776,"capability_primary_group_count":185,"capability_non_primary_group_count":252},{"model":"Claude Opus 4.8","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-05-29","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":526,"benchmark_group_count":440,"benchmark_coverage_score":265.0,"source_family_count":88,"source_group_family_count":86,"source_effective_breadth":5.65,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.4382,"source_primary_group_share":0.3795,"capability_breadth":3.91,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4554,"capability_primary_group_count":173,"capability_non_primary_group_count":265},{"model":"gemma-3-12B-pt","provider":"Google","country":"US","openness":"open","release_date":"2025-03-12","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":526,"benchmark_group_count":190,"benchmark_coverage_score":50.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.26,"source_primary":"euroeval.com","source_primary_share":0.5339,"source_primary_group_share":0.6263,"capability_breadth":2.22,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5955,"capability_primary_group_count":140,"capability_non_primary_group_count":50},{"model":"gemma-3-4B-pt Base","provider":"Google","country":"US","openness":"open","release_date":"2025-03-12","param_count":4.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":519,"benchmark_group_count":185,"benchmark_coverage_score":48.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":2.16,"source_primary":"euroeval.com","source_primary_share":0.5439,"source_primary_group_share":0.6432,"capability_breadth":2.22,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5941,"capability_primary_group_count":137,"capability_non_primary_group_count":48},{"model":"Qwen 2.5 1.5B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-19","param_count":1.5,"active_param_count":null,"training_compute_flop":1.663e+23,"num_benchmarks":506,"benchmark_group_count":242,"benchmark_coverage_score":80.0,"source_family_count":27,"source_group_family_count":27,"source_effective_breadth":4.57,"source_primary":"euroeval.com","source_primary_share":0.4663,"source_primary_group_share":0.4008,"capability_breadth":2.28,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5992,"capability_primary_group_count":162,"capability_non_primary_group_count":80},{"model":"Gemma 2 27B (IT)","provider":"Google","country":"US","openness":"open","release_date":"2024-06-27","param_count":27.0,"active_param_count":null,"training_compute_flop":2.106e+24,"num_benchmarks":505,"benchmark_group_count":324,"benchmark_coverage_score":178.0,"source_family_count":40,"source_group_family_count":39,"source_effective_breadth":9.82,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.245,"source_primary_group_share":0.1667,"capability_breadth":3.68,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4382,"capability_primary_group_count":146,"capability_non_primary_group_count":178},{"model":"GPT-5.1","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-11-12","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":500,"benchmark_group_count":419,"benchmark_coverage_score":236.0,"source_family_count":56,"source_group_family_count":56,"source_effective_breadth":4.95,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.4653,"source_primary_group_share":0.401,"capability_breadth":3.6,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5,"capability_primary_group_count":183,"capability_non_primary_group_count":236},{"model":"GPT-4.1 (2025-04-14)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-04-14","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":496,"benchmark_group_count":371,"benchmark_coverage_score":43.0,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":1.61,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.7213,"source_primary_group_share":0.7763,"capability_breadth":1.31,"capability_primary":"vision_language","capability_primary_share":0.8709,"capability_primary_group_count":328,"capability_non_primary_group_count":43},{"model":"Gemini 3 Pro (Preview)","provider":"Google","country":"US","openness":"api","release_date":"2025-11-18","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 3 Pro preview snapshot retired or superseded by newer Gemini models.","num_benchmarks":495,"benchmark_group_count":326,"benchmark_coverage_score":210.0,"source_family_count":66,"source_group_family_count":66,"source_effective_breadth":6.29,"source_primary":"euroeval.com","source_primary_share":0.5306,"source_primary_group_share":0.365,"capability_breadth":3.89,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4347,"capability_primary_group_count":116,"capability_non_primary_group_count":210},{"model":"Gemma 2 9B (IT)","provider":"Google","country":"US","openness":"open","release_date":"2024-06-27","param_count":9.0,"active_param_count":null,"training_compute_flop":4.32e+23,"num_benchmarks":488,"benchmark_group_count":296,"benchmark_coverage_score":142.0,"source_family_count":46,"source_group_family_count":46,"source_effective_breadth":12.01,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.2536,"source_primary_group_share":0.1419,"capability_breadth":3.05,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4907,"capability_primary_group_count":154,"capability_non_primary_group_count":142},{"model":"Claude 3 Haiku","provider":"Anthropic","country":"US","openness":"api","release_date":"2024-03-13","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 3 family retired or superseded by newer Claude models.","num_benchmarks":486,"benchmark_group_count":373,"benchmark_coverage_score":81.0,"source_family_count":39,"source_group_family_count":38,"source_effective_breadth":1.66,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.7214,"source_primary_group_share":0.7721,"capability_breadth":1.82,"capability_primary":"vision_language","capability_primary_share":0.7318,"capability_primary_group_count":292,"capability_non_primary_group_count":81},{"model":"Mistral Small 3.1","provider":"Mistral","country":"FR","openness":"open","release_date":"2025-03-17","param_count":24.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":485,"benchmark_group_count":298,"benchmark_coverage_score":181.0,"source_family_count":31,"source_group_family_count":31,"source_effective_breadth":4.57,"source_primary":"euroeval.com","source_primary_share":0.5439,"source_primary_group_share":0.3993,"capability_breadth":4.1,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4435,"capability_primary_group_count":117,"capability_non_primary_group_count":181},{"model":"Claude Opus 4.6","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-02-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":483,"benchmark_group_count":430,"benchmark_coverage_score":354.0,"source_family_count":117,"source_group_family_count":117,"source_effective_breadth":11.83,"source_primary":"benchmarklist.com","source_primary_share":0.2205,"source_primary_group_share":0.2326,"capability_breadth":8.2,"capability_primary":"agentic","capability_primary_share":0.1772,"capability_primary_group_count":75,"capability_non_primary_group_count":354},{"model":"Llama 3.2 1B","provider":"Meta","country":"US","openness":"open","release_date":"2024-09-25","param_count":1.0,"active_param_count":null,"training_compute_flop":6.642e+22,"num_benchmarks":474,"benchmark_group_count":216,"benchmark_coverage_score":53.0,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":2.75,"source_primary":"euroeval.com","source_primary_share":0.5508,"source_primary_group_share":0.5509,"capability_breadth":2.11,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6335,"capability_primary_group_count":163,"capability_non_primary_group_count":53},{"model":"Llama 3 8B Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2024-04-18","param_count":8.0,"active_param_count":null,"training_compute_flop":7.2e+23,"num_benchmarks":463,"benchmark_group_count":331,"benchmark_coverage_score":224.0,"source_family_count":60,"source_group_family_count":59,"source_effective_breadth":13.65,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.1429,"source_primary_group_share":0.1934,"capability_breadth":4.91,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3626,"capability_primary_group_count":107,"capability_non_primary_group_count":224},{"model":"Qwen 3 30B A3B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":30.0,"active_param_count":3.0,"training_compute_flop":6.48e+23,"num_benchmarks":456,"benchmark_group_count":302,"benchmark_coverage_score":78.0,"source_family_count":41,"source_group_family_count":41,"source_effective_breadth":2.77,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.5055,"source_primary_group_share":0.5563,"capability_breadth":1.78,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7406,"capability_primary_group_count":224,"capability_non_primary_group_count":78},{"model":"GPT-OSS-20B","provider":"OpenAI","country":"US","openness":"open","release_date":"2025-08-05","param_count":20.0,"active_param_count":3.61,"training_compute_flop":5.49e+23,"num_benchmarks":454,"benchmark_group_count":348,"benchmark_coverage_score":159.0,"source_family_count":56,"source_group_family_count":55,"source_effective_breadth":3.91,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.5078,"source_primary_group_share":0.4828,"capability_breadth":2.8,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5768,"capability_primary_group_count":189,"capability_non_primary_group_count":159},{"model":"DeepSeek V3.1","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-08-21","param_count":671.0,"active_param_count":37.0,"training_compute_flop":3.594058e+24,"num_benchmarks":453,"benchmark_group_count":353,"benchmark_coverage_score":166.0,"source_family_count":48,"source_group_family_count":48,"source_effective_breadth":3.77,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.5158,"source_primary_group_share":0.4759,"capability_breadth":2.99,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5588,"capability_primary_group_count":187,"capability_non_primary_group_count":166},{"model":"Phi-4 Mini Instruct","provider":"Microsoft","country":"US","openness":"open","release_date":"2025-02-26","param_count":3.8,"active_param_count":null,"training_compute_flop":9.956e+22,"num_benchmarks":453,"benchmark_group_count":276,"benchmark_coverage_score":80.0,"source_family_count":19,"source_group_family_count":19,"source_effective_breadth":2.52,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.5044,"source_primary_group_share":0.6087,"capability_breadth":2.18,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6327,"capability_primary_group_count":196,"capability_non_primary_group_count":80},{"model":"Llama 3.2 90B Vision Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2024-09-25","param_count":90.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":450,"benchmark_group_count":377,"benchmark_coverage_score":63.0,"source_family_count":18,"source_group_family_count":18,"source_effective_breadth":1.66,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.7843,"source_primary_group_share":0.7692,"capability_breadth":1.39,"capability_primary":"vision_language","capability_primary_share":0.8449,"capability_primary_group_count":314,"capability_non_primary_group_count":63},{"model":"Kimi K2","provider":"Moonshot","country":"CN","openness":"open","release_date":"2025-07-11","param_count":1000.0,"active_param_count":32.0,"training_compute_flop":2.976e+24,"num_benchmarks":448,"benchmark_group_count":345,"benchmark_coverage_score":172.0,"source_family_count":76,"source_group_family_count":76,"source_effective_breadth":4.07,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.5124,"source_primary_group_share":0.487,"capability_breadth":3.24,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5236,"capability_primary_group_count":173,"capability_non_primary_group_count":172},{"model":"Qwen 3 14B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":14.0,"active_param_count":null,"training_compute_flop":3.197e+24,"num_benchmarks":444,"benchmark_group_count":260,"benchmark_coverage_score":168.0,"source_family_count":42,"source_group_family_count":42,"source_effective_breadth":5.97,"source_primary":"euroeval.com","source_primary_share":0.3032,"source_primary_group_share":0.25,"capability_breadth":3.96,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3889,"capability_primary_group_count":92,"capability_non_primary_group_count":168},{"model":"Llama 3.2 1B Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2024-09-25","param_count":1.0,"active_param_count":null,"training_compute_flop":6.642e+22,"num_benchmarks":438,"benchmark_group_count":256,"benchmark_coverage_score":89.0,"source_family_count":43,"source_group_family_count":43,"source_effective_breadth":4.07,"source_primary":"euroeval.com","source_primary_share":0.5977,"source_primary_group_share":0.4648,"capability_breadth":2.14,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6598,"capability_primary_group_count":167,"capability_non_primary_group_count":89},{"model":"Gemma 4 31B (IT)","provider":"Google","country":"US","openness":"open","release_date":"2026-04-02","param_count":30.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":437,"benchmark_group_count":264,"benchmark_coverage_score":147.0,"source_family_count":37,"source_group_family_count":36,"source_effective_breadth":3.7,"source_primary":"euroeval.com","source_primary_share":0.6032,"source_primary_group_share":0.4508,"capability_breadth":3.53,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4919,"capability_primary_group_count":117,"capability_non_primary_group_count":147},{"model":"Claude Sonnet 4.5 (Thinking)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-09-29","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":435,"benchmark_group_count":238,"benchmark_coverage_score":130.0,"source_family_count":23,"source_group_family_count":23,"source_effective_breadth":3.61,"source_primary":"euroeval.com","source_primary_share":0.6061,"source_primary_group_share":0.5,"capability_breadth":3.37,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4965,"capability_primary_group_count":108,"capability_non_primary_group_count":130},{"model":"GPT-3.5 Turbo","provider":"OpenAI","country":"US","openness":"api","release_date":"2022-11-30","param_count":175.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI GPT-3/GPT-3.5 family retired or superseded by newer API models.","num_benchmarks":422,"benchmark_group_count":318,"benchmark_coverage_score":135.0,"source_family_count":46,"source_group_family_count":46,"source_effective_breadth":3.44,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.5602,"source_primary_group_share":0.5283,"capability_breadth":2.56,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6044,"capability_primary_group_count":183,"capability_non_primary_group_count":135},{"model":"Grok 4.1 Fast (Reasoning)","provider":"xAI","country":"US","openness":"api","release_date":"2025-11-20","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":422,"benchmark_group_count":233,"benchmark_coverage_score":126.0,"source_family_count":31,"source_group_family_count":31,"source_effective_breadth":3.46,"source_primary":"euroeval.com","source_primary_share":0.6235,"source_primary_group_share":0.5107,"capability_breadth":3.29,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5084,"capability_primary_group_count":107,"capability_non_primary_group_count":126},{"model":"Qwen 3 32B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":32.0,"active_param_count":null,"training_compute_flop":7.085e+24,"num_benchmarks":414,"benchmark_group_count":247,"benchmark_coverage_score":174.0,"source_family_count":57,"source_group_family_count":57,"source_effective_breadth":7.29,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.303,"source_primary_group_share":0.2632,"capability_breadth":4.42,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3227,"capability_primary_group_count":73,"capability_non_primary_group_count":174},{"model":"Llama 3.1 70B","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-23","param_count":70.0,"active_param_count":null,"training_compute_flop":7.929e+24,"num_benchmarks":414,"benchmark_group_count":240,"benchmark_coverage_score":81.0,"source_family_count":27,"source_group_family_count":27,"source_effective_breadth":3.44,"source_primary":"euroeval.com","source_primary_share":0.6357,"source_primary_group_share":0.4958,"capability_breadth":2.12,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6626,"capability_primary_group_count":159,"capability_non_primary_group_count":81},{"model":"Qwen 3.5 9B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-03-02","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":412,"benchmark_group_count":243,"benchmark_coverage_score":131.0,"source_family_count":31,"source_group_family_count":30,"source_effective_breadth":3.48,"source_primary":"euroeval.com","source_primary_share":0.6373,"source_primary_group_share":0.4897,"capability_breadth":3.21,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5196,"capability_primary_group_count":112,"capability_non_primary_group_count":131},{"model":"DeepSeek V3.2","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-12-01","param_count":671.0,"active_param_count":37.0,"training_compute_flop":4.2e+24,"num_benchmarks":410,"benchmark_group_count":358,"benchmark_coverage_score":284.0,"source_family_count":85,"source_group_family_count":83,"source_effective_breadth":12.4,"source_primary":"benchmarklist.com","source_primary_share":0.1722,"source_primary_group_share":0.1872,"capability_breadth":8.54,"capability_primary":"reasoning_logical","capability_primary_share":0.2108,"capability_primary_group_count":74,"capability_non_primary_group_count":284},{"model":"QwQ-32B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-03-05","param_count":32.0,"active_param_count":null,"training_compute_flop":3.51e+24,"num_benchmarks":409,"benchmark_group_count":229,"benchmark_coverage_score":135.0,"source_family_count":44,"source_group_family_count":44,"source_effective_breadth":8.89,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.3015,"source_primary_group_share":0.2183,"capability_breadth":3.77,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4044,"capability_primary_group_count":94,"capability_non_primary_group_count":135},{"model":"Claude 3 Sonnet","provider":"Anthropic","country":"US","openness":"api","release_date":"2024-03-04","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 3 family retired or superseded by newer Claude models.","num_benchmarks":406,"benchmark_group_count":336,"benchmark_coverage_score":44.0,"source_family_count":24,"source_group_family_count":23,"source_effective_breadth":1.36,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.8675,"source_primary_group_share":0.8571,"capability_breadth":1.29,"capability_primary":"vision_language","capability_primary_share":0.8775,"capability_primary_group_count":292,"capability_non_primary_group_count":44},{"model":"Olmo 3.1 32B Instruct","provider":"Allen AI","country":"US","openness":"open","release_date":"2025-12-23","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":406,"benchmark_group_count":237,"benchmark_coverage_score":129.0,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":2.95,"source_primary":"euroeval.com","source_primary_share":0.6468,"source_primary_group_share":0.5021,"capability_breadth":3.24,"capability_primary":"knowledge_multilingual","capability_primary_share":0.505,"capability_primary_group_count":108,"capability_non_primary_group_count":129},{"model":"Mistral Medium 3.1","provider":"Mistral","country":"FR","openness":"api","release_date":"2025-08-13","param_count":123.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":404,"benchmark_group_count":301,"benchmark_coverage_score":125.0,"source_family_count":30,"source_group_family_count":30,"source_effective_breadth":2.75,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.5743,"source_primary_group_share":0.5581,"capability_breadth":2.63,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5995,"capability_primary_group_count":176,"capability_non_primary_group_count":125},{"model":"Gemini 3.1 Flash Lite (Preview)","provider":"Google","country":"US","openness":"api","release_date":"2026-05-07","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Google lists gemini-3.1-flash-lite-preview with a May 25, 2026 shutdown date and recommends gemini-3.1-flash-lite.","num_benchmarks":403,"benchmark_group_count":244,"benchmark_coverage_score":134.0,"source_family_count":38,"source_group_family_count":38,"source_effective_breadth":3.87,"source_primary":"euroeval.com","source_primary_share":0.6516,"source_primary_group_share":0.4877,"capability_breadth":3.26,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5188,"capability_primary_group_count":110,"capability_non_primary_group_count":134},{"model":"Gemma 4 31B","provider":"Google","country":"US","openness":"open","release_date":"2026-04-02","param_count":30.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":402,"benchmark_group_count":240,"benchmark_coverage_score":126.0,"source_family_count":32,"source_group_family_count":31,"source_effective_breadth":3.14,"source_primary":"euroeval.com","source_primary_share":0.6549,"source_primary_group_share":0.4958,"capability_breadth":3.08,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5328,"capability_primary_group_count":113,"capability_non_primary_group_count":126},{"model":"Grok 4","provider":"xAI","country":"US","openness":"api","release_date":"2025-07-09","param_count":null,"active_param_count":null,"training_compute_flop":5e+26,"num_benchmarks":399,"benchmark_group_count":317,"benchmark_coverage_score":240.0,"source_family_count":101,"source_group_family_count":101,"source_effective_breadth":28.8,"source_primary":"artificialanalysis.ai","source_primary_share":0.1362,"source_primary_group_share":0.1009,"capability_breadth":7.51,"capability_primary":"knowledge_general","capability_primary_share":0.2596,"capability_primary_group_count":77,"capability_non_primary_group_count":240},{"model":"gemma-3-27B-pt","provider":"Google","country":"US","openness":"open","release_date":"2025-03-12","param_count":27.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":395,"benchmark_group_count":158,"benchmark_coverage_score":36.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.61,"source_primary":"euroeval.com","source_primary_share":0.6599,"source_primary_group_share":0.7532,"capability_breadth":2.11,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6294,"capability_primary_group_count":122,"capability_non_primary_group_count":36},{"model":"GPT-OSS-20B (High)","provider":"OpenAI","country":"US","openness":"open","release_date":"2025-08-05","param_count":20.0,"active_param_count":3.61,"training_compute_flop":5.49e+23,"num_benchmarks":394,"benchmark_group_count":204,"benchmark_coverage_score":96.0,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":2.61,"source_primary":"euroeval.com","source_primary_share":0.6616,"source_primary_group_share":0.5833,"capability_breadth":2.77,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5408,"capability_primary_group_count":107,"capability_non_primary_group_count":96},{"model":"Qwen 2.5 7B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-19","param_count":7.0,"active_param_count":null,"training_compute_flop":8.219e+23,"num_benchmarks":393,"benchmark_group_count":221,"benchmark_coverage_score":126.0,"source_family_count":53,"source_group_family_count":52,"source_effective_breadth":11.7,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.3162,"source_primary_group_share":0.1674,"capability_breadth":3.56,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4113,"capability_primary_group_count":95,"capability_non_primary_group_count":126},{"model":"Gemma 4 26B A4B (IT)","provider":"Google","country":"US","openness":"open","release_date":"2026-04-02","param_count":25.2,"active_param_count":3.8,"training_compute_flop":null,"num_benchmarks":388,"benchmark_group_count":220,"benchmark_coverage_score":104.0,"source_family_count":22,"source_group_family_count":22,"source_effective_breadth":2.62,"source_primary":"euroeval.com","source_primary_share":0.6824,"source_primary_group_share":0.5409,"capability_breadth":2.9,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5538,"capability_primary_group_count":116,"capability_non_primary_group_count":104},{"model":"Gemma 3 1B (IT)","provider":"Google","country":"US","openness":"open","release_date":"2025-03-12","param_count":1.0,"active_param_count":null,"training_compute_flop":1.2e+22,"num_benchmarks":381,"benchmark_group_count":185,"benchmark_coverage_score":59.0,"source_family_count":19,"source_group_family_count":19,"source_effective_breadth":2.31,"source_primary":"euroeval.com","source_primary_share":0.6842,"source_primary_group_share":0.6432,"capability_breadth":2.23,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6421,"capability_primary_group_count":126,"capability_non_primary_group_count":59},{"model":"Kimi K2.5","provider":"Moonshot","country":"CN","openness":"open","release_date":"2026-01-27","param_count":1000.0,"active_param_count":32.0,"training_compute_flop":5.8e+24,"num_benchmarks":374,"benchmark_group_count":296,"benchmark_coverage_score":247.0,"source_family_count":93,"source_group_family_count":90,"source_effective_breadth":15.04,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.1838,"source_primary_group_share":0.2196,"capability_breadth":8.66,"capability_primary":"code","capability_primary_share":0.1838,"capability_primary_group_count":49,"capability_non_primary_group_count":247},{"model":"GLM 4.5 Air","provider":"Zhipu","country":"CN","openness":"open","release_date":"2025-07-28","param_count":106.0,"active_param_count":12.0,"training_compute_flop":1.656e+24,"num_benchmarks":374,"benchmark_group_count":269,"benchmark_coverage_score":101.0,"source_family_count":30,"source_group_family_count":30,"source_effective_breadth":2.48,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.6162,"source_primary_group_share":0.6245,"capability_breadth":2.43,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6162,"capability_primary_group_count":168,"capability_non_primary_group_count":101},{"model":"Gemini 3.1 Flash Lite","provider":"Google","country":"US","openness":"api","release_date":"2026-05-07","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":368,"benchmark_group_count":264,"benchmark_coverage_score":94.0,"source_family_count":36,"source_group_family_count":34,"source_effective_breadth":2.39,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.6196,"source_primary_group_share":0.6364,"capability_breadth":2.41,"capability_primary":"knowledge_multilingual","capability_primary_share":0.624,"capability_primary_group_count":169,"capability_non_primary_group_count":94},{"model":"Gemma 2 2B (IT)","provider":"Google","country":"US","openness":"open","release_date":"2024-06-27","param_count":2.0,"active_param_count":null,"training_compute_flop":3.12e+22,"num_benchmarks":365,"benchmark_group_count":189,"benchmark_coverage_score":80.0,"source_family_count":28,"source_group_family_count":28,"source_effective_breadth":6.91,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.3388,"source_primary_group_share":0.2222,"capability_breadth":2.72,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5041,"capability_primary_group_count":109,"capability_non_primary_group_count":80},{"model":"OLMo 3.1 32B (Thinking)","provider":"Allen AI","country":"US","openness":"open","release_date":"2026-01-15","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":365,"benchmark_group_count":180,"benchmark_coverage_score":73.0,"source_family_count":15,"source_group_family_count":15,"source_effective_breadth":2.15,"source_primary":"euroeval.com","source_primary_share":0.7143,"source_primary_group_share":0.6611,"capability_breadth":2.57,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5824,"capability_primary_group_count":107,"capability_non_primary_group_count":73},{"model":"GLM-5.1","provider":"Zhipu","country":"CN","openness":"open","release_date":"2026-03-27","param_count":744.0,"active_param_count":40.0,"training_compute_flop":null,"num_benchmarks":364,"benchmark_group_count":312,"benchmark_coverage_score":243.0,"source_family_count":79,"source_group_family_count":77,"source_effective_breadth":9.02,"source_primary":"benchmarklist.com","source_primary_share":0.212,"source_primary_group_share":0.234,"capability_breadth":7.78,"capability_primary":"code","capability_primary_share":0.2174,"capability_primary_group_count":65,"capability_non_primary_group_count":243},{"model":"Qwen 2.5 32B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-19","param_count":32.0,"active_param_count":null,"training_compute_flop":3.51e+24,"num_benchmarks":363,"benchmark_group_count":209,"benchmark_coverage_score":132.0,"source_family_count":39,"source_group_family_count":38,"source_effective_breadth":10.59,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.3407,"source_primary_group_share":0.177,"capability_breadth":3.5,"capability_primary":"knowledge_general","capability_primary_share":0.3767,"capability_primary_group_count":77,"capability_non_primary_group_count":132},{"model":"Kimi K2.6","provider":"Moonshot","country":"CN","openness":"open","release_date":"2026-04-20","param_count":1000.0,"active_param_count":32.0,"training_compute_flop":null,"num_benchmarks":363,"benchmark_group_count":287,"benchmark_coverage_score":226.0,"source_family_count":89,"source_group_family_count":86,"source_effective_breadth":13.02,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.1886,"source_primary_group_share":0.2265,"capability_breadth":7.67,"capability_primary":"code","capability_primary_share":0.2254,"capability_primary_group_count":57,"capability_non_primary_group_count":226},{"model":"Qwen 3 4B 2507 Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-07-01","param_count":4.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":357,"benchmark_group_count":153,"benchmark_coverage_score":73.0,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":3.71,"source_primary":"euroeval.com","source_primary_share":0.3904,"source_primary_group_share":0.4379,"capability_breadth":2.84,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4719,"capability_primary_group_count":80,"capability_non_primary_group_count":73},{"model":"aya-expanse-32B","provider":"Cohere","country":"CA","openness":"open","release_date":"2024-10-24","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":354,"benchmark_group_count":183,"benchmark_coverage_score":64.0,"source_family_count":19,"source_group_family_count":19,"source_effective_breadth":6.0,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.3475,"source_primary_group_share":0.2732,"capability_breadth":2.49,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5537,"capability_primary_group_count":119,"capability_non_primary_group_count":64},{"model":"Gemini 2.5 Flash Lite","provider":"Google","country":"US","openness":"api","release_date":"2025-06-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":353,"benchmark_group_count":263,"benchmark_coverage_score":80.0,"source_family_count":33,"source_group_family_count":33,"source_effective_breadth":2.39,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.6647,"source_primary_group_share":0.6388,"capability_breadth":1.91,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7143,"capability_primary_group_count":183,"capability_non_primary_group_count":80},{"model":"DeepSeek R1 Distill Qwen 32B","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-01-20","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":349,"benchmark_group_count":186,"benchmark_coverage_score":92.0,"source_family_count":29,"source_group_family_count":29,"source_effective_breadth":6.35,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.3545,"source_primary_group_share":0.2688,"capability_breadth":3.17,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4611,"capability_primary_group_count":94,"capability_non_primary_group_count":92},{"model":"DeepSeek R1 Distill Qwen 14B","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-01-20","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":347,"benchmark_group_count":178,"benchmark_coverage_score":85.0,"source_family_count":23,"source_group_family_count":23,"source_effective_breadth":5.81,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.3555,"source_primary_group_share":0.2809,"capability_breadth":3.16,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4538,"capability_primary_group_count":93,"capability_non_primary_group_count":85},{"model":"GPT-OSS-20B (Low)","provider":"OpenAI","country":"US","openness":"open","release_date":"2025-08-05","param_count":20.0,"active_param_count":3.61,"training_compute_flop":5.49e+23,"num_benchmarks":345,"benchmark_group_count":166,"benchmark_coverage_score":59.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":1.83,"source_primary":"euroeval.com","source_primary_share":0.7558,"source_primary_group_share":0.7169,"capability_breadth":2.31,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6163,"capability_primary_group_count":107,"capability_non_primary_group_count":59},{"model":"Qwen 3.5 4B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-03-01","param_count":4.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":345,"benchmark_group_count":186,"benchmark_coverage_score":75.0,"source_family_count":16,"source_group_family_count":15,"source_effective_breadth":2.33,"source_primary":"euroeval.com","source_primary_share":0.7558,"source_primary_group_share":0.6398,"capability_breadth":2.43,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6134,"capability_primary_group_count":111,"capability_non_primary_group_count":75},{"model":"Qwen 3.5 2B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-03-02","param_count":2.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":345,"benchmark_group_count":179,"benchmark_coverage_score":69.0,"source_family_count":14,"source_group_family_count":13,"source_effective_breadth":2.18,"source_primary":"euroeval.com","source_primary_share":0.7558,"source_primary_group_share":0.6648,"capability_breadth":2.45,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6105,"capability_primary_group_count":110,"capability_non_primary_group_count":69},{"model":"Qwen 3 0.6B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":0.6,"active_param_count":null,"training_compute_flop":1.296e+23,"num_benchmarks":343,"benchmark_group_count":141,"benchmark_coverage_score":47.0,"source_family_count":18,"source_group_family_count":18,"source_effective_breadth":2.19,"source_primary":"euroeval.com","source_primary_share":0.5455,"source_primary_group_share":0.6241,"capability_breadth":2.37,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5601,"capability_primary_group_count":94,"capability_non_primary_group_count":47},{"model":"Mistral Nemo","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-07-18","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":341,"benchmark_group_count":268,"benchmark_coverage_score":94.0,"source_family_count":18,"source_group_family_count":18,"source_effective_breadth":2.21,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.6786,"source_primary_group_share":0.6269,"capability_breadth":2.01,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6964,"capability_primary_group_count":174,"capability_non_primary_group_count":94},{"model":"DeepSeek V3.2 Exp","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-09-29","param_count":671.0,"active_param_count":37.0,"training_compute_flop":4.18e+24,"num_benchmarks":341,"benchmark_group_count":272,"benchmark_coverage_score":99.0,"source_family_count":26,"source_group_family_count":26,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.6806,"source_primary_group_share":0.6176,"capability_breadth":2.01,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6955,"capability_primary_group_count":173,"capability_non_primary_group_count":99},{"model":"DeepSeek R1","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-01-20","param_count":671.0,"active_param_count":37.0,"training_compute_flop":4.02001e+24,"num_benchmarks":337,"benchmark_group_count":278,"benchmark_coverage_score":198.0,"source_family_count":112,"source_group_family_count":111,"source_effective_breadth":34.94,"source_primary":"crfm.stanford.edu","source_primary_share":0.1169,"source_primary_group_share":0.1259,"capability_breadth":6.79,"capability_primary":"knowledge_general","capability_primary_share":0.2831,"capability_primary_group_count":80,"capability_non_primary_group_count":198},{"model":"Qwen 3 Next 80B A3B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-09-11","param_count":80.0,"active_param_count":3.0,"training_compute_flop":2.7e+23,"num_benchmarks":337,"benchmark_group_count":232,"benchmark_coverage_score":165.0,"source_family_count":25,"source_group_family_count":24,"source_effective_breadth":5.61,"source_primary":"euroeval.com","source_primary_share":0.3958,"source_primary_group_share":0.2802,"capability_breadth":4.99,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3353,"capability_primary_group_count":67,"capability_non_primary_group_count":165},{"model":"Ministral 3 3B","provider":"Mistral","country":"FR","openness":"open","release_date":"2025-12-02","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":336,"benchmark_group_count":154,"benchmark_coverage_score":46.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":1.62,"source_primary":"euroeval.com","source_primary_share":0.7738,"source_primary_group_share":0.7727,"capability_breadth":2.27,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6339,"capability_primary_group_count":108,"capability_non_primary_group_count":46},{"model":"Grok 4.20","provider":"xAI","country":"US","openness":"api","release_date":"2026-03-10","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":335,"benchmark_group_count":255,"benchmark_coverage_score":88.0,"source_family_count":35,"source_group_family_count":34,"source_effective_breadth":2.18,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.7015,"source_primary_group_share":0.6588,"capability_breadth":2.0,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6985,"capability_primary_group_count":167,"capability_non_primary_group_count":88},{"model":"Nova Pro (v1)","provider":"Amazon","country":"US","openness":"api","release_date":"2024-12-03","param_count":90.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":335,"benchmark_group_count":261,"benchmark_coverage_score":84.0,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":2.09,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.693,"source_primary_group_share":0.6437,"capability_breadth":1.88,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7204,"capability_primary_group_count":177,"capability_non_primary_group_count":84},{"model":"OLMo 3 7B Instruct","provider":"Allen AI","country":"US","openness":"open","release_date":"2025-11-20","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":333,"benchmark_group_count":168,"benchmark_coverage_score":64.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":1.89,"source_primary":"euroeval.com","source_primary_share":0.7831,"source_primary_group_share":0.7083,"capability_breadth":2.42,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5994,"capability_primary_group_count":104,"capability_non_primary_group_count":64},{"model":"OLMo 3 7B (Thinking)","provider":"Allen AI","country":"US","openness":"open","release_date":"2025-11-20","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":332,"benchmark_group_count":167,"benchmark_coverage_score":64.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":1.87,"source_primary":"euroeval.com","source_primary_share":0.7855,"source_primary_group_share":0.7126,"capability_breadth":2.42,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5982,"capability_primary_group_count":103,"capability_non_primary_group_count":64},{"model":"Gemma 4 26B A4B","provider":"Google","country":"US","openness":"open","release_date":"2026-04-02","param_count":25.2,"active_param_count":3.8,"training_compute_flop":null,"num_benchmarks":332,"benchmark_group_count":176,"benchmark_coverage_score":68.0,"source_family_count":15,"source_group_family_count":14,"source_effective_breadth":2.05,"source_primary":"euroeval.com","source_primary_share":0.7903,"source_primary_group_share":0.6761,"capability_breadth":2.37,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6261,"capability_primary_group_count":108,"capability_non_primary_group_count":68},{"model":"Qwen 3 4B 2507 (Thinking)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-07-01","param_count":4.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":331,"benchmark_group_count":163,"benchmark_coverage_score":58.0,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":1.82,"source_primary":"euroeval.com","source_primary_share":0.7879,"source_primary_group_share":0.7301,"capability_breadth":2.36,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6182,"capability_primary_group_count":105,"capability_non_primary_group_count":58},{"model":"aya-23-35B","provider":"Cohere","country":"CA","openness":"open","release_date":"2024-05-23","param_count":35.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":328,"benchmark_group_count":154,"benchmark_coverage_score":50.0,"source_family_count":14,"source_group_family_count":14,"source_effective_breadth":4.47,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.375,"source_primary_group_share":0.3247,"capability_breadth":2.42,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5488,"capability_primary_group_count":104,"capability_non_primary_group_count":50},{"model":"nova-2-lite-v1","provider":"Amazon","country":"US","openness":"api","release_date":"2025-12-02","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":325,"benchmark_group_count":250,"benchmark_coverage_score":79.0,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":1.94,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.7138,"source_primary_group_share":0.668,"capability_breadth":1.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7264,"capability_primary_group_count":171,"capability_non_primary_group_count":79},{"model":"jamba-large-1.7","provider":"AI21 Labs","country":"IL","openness":"api","release_date":"2025-07-01","param_count":398.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":323,"benchmark_group_count":250,"benchmark_coverage_score":74.0,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":1.93,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.717,"source_primary_group_share":0.672,"capability_breadth":1.78,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7421,"capability_primary_group_count":176,"capability_non_primary_group_count":74},{"model":"aya-expanse-8B","provider":"Cohere","country":"CA","openness":"open","release_date":"2024-10-24","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":322,"benchmark_group_count":171,"benchmark_coverage_score":56.0,"source_family_count":19,"source_group_family_count":19,"source_effective_breadth":6.68,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.382,"source_primary_group_share":0.2164,"capability_breadth":2.35,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5652,"capability_primary_group_count":115,"capability_non_primary_group_count":56},{"model":"SOLAR-10.7B-v1.0","provider":"Upstage","country":"KR","openness":"open","release_date":"2023-12-11","param_count":10.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":321,"benchmark_group_count":123,"benchmark_coverage_score":48.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.57,"source_primary":"euroeval.com","source_primary_share":0.5389,"source_primary_group_share":0.5447,"capability_breadth":2.43,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5421,"capability_primary_group_count":75,"capability_non_primary_group_count":48},{"model":"Qwen 2 7B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-06-07","param_count":7.0,"active_param_count":null,"training_compute_flop":2.94e+23,"num_benchmarks":320,"benchmark_group_count":171,"benchmark_coverage_score":82.0,"source_family_count":29,"source_group_family_count":28,"source_effective_breadth":7.86,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.3844,"source_primary_group_share":0.2164,"capability_breadth":2.88,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4844,"capability_primary_group_count":89,"capability_non_primary_group_count":82},{"model":"Qwen 3 4B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":4.0,"active_param_count":null,"training_compute_flop":8.64e+23,"num_benchmarks":320,"benchmark_group_count":150,"benchmark_coverage_score":72.0,"source_family_count":30,"source_group_family_count":30,"source_effective_breadth":3.92,"source_primary":"euroeval.com","source_primary_share":0.4107,"source_primary_group_share":0.4333,"capability_breadth":2.78,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4765,"capability_primary_group_count":78,"capability_non_primary_group_count":72},{"model":"GPT-5.4 Mini (Medium)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-03-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":320,"benchmark_group_count":148,"benchmark_coverage_score":45.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":1.52,"source_primary":"euroeval.com","source_primary_share":0.8125,"source_primary_group_share":0.8041,"capability_breadth":2.38,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6188,"capability_primary_group_count":103,"capability_non_primary_group_count":45},{"model":"Command A","provider":"Cohere","country":"CA","openness":"open","release_date":"2025-03-13","param_count":111.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":318,"benchmark_group_count":231,"benchmark_coverage_score":65.0,"source_family_count":17,"source_group_family_count":17,"source_effective_breadth":1.85,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.7238,"source_primary_group_share":0.7273,"capability_breadth":1.87,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7175,"capability_primary_group_count":166,"capability_non_primary_group_count":65},{"model":"GPT-5.4 Nano (Medium)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-03-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":318,"benchmark_group_count":146,"benchmark_coverage_score":43.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":1.48,"source_primary":"euroeval.com","source_primary_share":0.8176,"source_primary_group_share":0.8151,"capability_breadth":2.36,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6226,"capability_primary_group_count":103,"capability_non_primary_group_count":43},{"model":"Tiny Aya Global","provider":"Cohere","country":"CA","openness":"open","release_date":"2026-02-17","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":316,"benchmark_group_count":141,"benchmark_coverage_score":34.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.37,"source_primary":"euroeval.com","source_primary_share":0.8228,"source_primary_group_share":0.844,"capability_breadth":2.08,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6709,"capability_primary_group_count":107,"capability_non_primary_group_count":34},{"model":"Qwen 3 1.7B (Non-reasoning)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":1.7,"active_param_count":null,"training_compute_flop":3.672e+23,"num_benchmarks":316,"benchmark_group_count":144,"benchmark_coverage_score":38.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.42,"source_primary":"euroeval.com","source_primary_share":0.8254,"source_primary_group_share":0.8264,"capability_breadth":2.09,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6667,"capability_primary_group_count":106,"capability_non_primary_group_count":38},{"model":"Qwen 3.5 0.8B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-03-02","param_count":0.8,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":316,"benchmark_group_count":159,"benchmark_coverage_score":49.0,"source_family_count":12,"source_group_family_count":11,"source_effective_breadth":1.74,"source_primary":"euroeval.com","source_primary_share":0.8228,"source_primary_group_share":0.7484,"capability_breadth":2.13,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6646,"capability_primary_group_count":110,"capability_non_primary_group_count":49},{"model":"Qwen 3 Next 80B A3B (Thinking)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-09-11","param_count":80.0,"active_param_count":3.0,"training_compute_flop":2.7e+23,"num_benchmarks":315,"benchmark_group_count":185,"benchmark_coverage_score":129.0,"source_family_count":29,"source_group_family_count":28,"source_effective_breadth":6.05,"source_primary":"euroeval.com","source_primary_share":0.4304,"source_primary_group_share":0.3514,"capability_breadth":4.44,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3625,"capability_primary_group_count":56,"capability_non_primary_group_count":129},{"model":"Gemma 4 E4B","provider":"Google","country":"US","openness":"open","release_date":"2026-04-02","param_count":4.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":312,"benchmark_group_count":159,"benchmark_coverage_score":52.0,"source_family_count":12,"source_group_family_count":11,"source_effective_breadth":1.73,"source_primary":"euroeval.com","source_primary_share":0.8387,"source_primary_group_share":0.7484,"capability_breadth":2.14,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6548,"capability_primary_group_count":107,"capability_non_primary_group_count":52},{"model":"gemma-4-E4B-it","provider":"Google","country":"US","openness":"open","release_date":"2026-04-02","param_count":4.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":311,"benchmark_group_count":155,"benchmark_coverage_score":47.0,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":1.66,"source_primary":"euroeval.com","source_primary_share":0.8387,"source_primary_group_share":0.7677,"capability_breadth":2.16,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6548,"capability_primary_group_count":108,"capability_non_primary_group_count":47},{"model":"GPT-5.4 Mini (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-03-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":308,"benchmark_group_count":158,"benchmark_coverage_score":52.0,"source_family_count":14,"source_group_family_count":14,"source_effective_breadth":1.73,"source_primary":"euroeval.com","source_primary_share":0.8442,"source_primary_group_share":0.7532,"capability_breadth":2.18,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6526,"capability_primary_group_count":106,"capability_non_primary_group_count":52},{"model":"Gemma 3 270M","provider":"Google","country":"US","openness":"open","release_date":"2025-03-12","param_count":0.27,"active_param_count":null,"training_compute_flop":9.72e+21,"num_benchmarks":308,"benchmark_group_count":149,"benchmark_coverage_score":39.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.52,"source_primary":"euroeval.com","source_primary_share":0.8469,"source_primary_group_share":0.7987,"capability_breadth":2.09,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6678,"capability_primary_group_count":110,"capability_non_primary_group_count":39},{"model":"Gemma 3n E4B (IT)","provider":"Google","country":"US","openness":"open","release_date":"2025-06-26","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":308,"benchmark_group_count":160,"benchmark_coverage_score":36.0,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":1.73,"source_primary":"euroeval.com","source_primary_share":0.8442,"source_primary_group_share":0.7438,"capability_breadth":1.78,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7305,"capability_primary_group_count":124,"capability_non_primary_group_count":36},{"model":"Claude Sonnet 4.5 (Non-reasoning)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-09-29","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":306,"benchmark_group_count":146,"benchmark_coverage_score":43.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":1.47,"source_primary":"euroeval.com","source_primary_share":0.8525,"source_primary_group_share":0.8151,"capability_breadth":2.2,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6492,"capability_primary_group_count":103,"capability_non_primary_group_count":43},{"model":"LFM2 1.2B","provider":"Liquid AI","country":"US","openness":"open","release_date":"2025-11-28","param_count":1.2,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":304,"benchmark_group_count":139,"benchmark_coverage_score":36.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.34,"source_primary":"euroeval.com","source_primary_share":0.8553,"source_primary_group_share":0.8561,"capability_breadth":2.17,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6513,"capability_primary_group_count":103,"capability_non_primary_group_count":36},{"model":"Gemini 2.5 Flash (Thinking)","provider":"Google","country":"US","openness":"api","release_date":"2025-06-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":303,"benchmark_group_count":143,"benchmark_coverage_score":61.0,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":2.06,"source_primary":"euroeval.com","source_primary_share":0.7781,"source_primary_group_share":0.6783,"capability_breadth":2.66,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5762,"capability_primary_group_count":82,"capability_non_primary_group_count":61},{"model":"Gemma 4 E2B","provider":"Google","country":"US","openness":"open","release_date":"2026-04-02","param_count":2.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":303,"benchmark_group_count":152,"benchmark_coverage_score":46.0,"source_family_count":11,"source_group_family_count":10,"source_effective_breadth":1.6,"source_primary":"euroeval.com","source_primary_share":0.8581,"source_primary_group_share":0.7829,"capability_breadth":2.07,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6667,"capability_primary_group_count":106,"capability_non_primary_group_count":46},{"model":"Claude Haiku 4.5 (20251001)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-10-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":303,"benchmark_group_count":159,"benchmark_coverage_score":48.0,"source_family_count":20,"source_group_family_count":20,"source_effective_breadth":1.77,"source_primary":"euroeval.com","source_primary_share":0.8609,"source_primary_group_share":0.7484,"capability_breadth":2.03,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6821,"capability_primary_group_count":111,"capability_non_primary_group_count":48},{"model":"LFM2 2.6B","provider":"Liquid AI","country":"US","openness":"open","release_date":"2025-11-28","param_count":2.6,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":301,"benchmark_group_count":124,"benchmark_coverage_score":37.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":1.53,"source_primary":"euroeval.com","source_primary_share":0.7867,"source_primary_group_share":0.7903,"capability_breadth":2.28,"capability_primary":"knowledge_multilingual","capability_primary_share":0.63,"capability_primary_group_count":87,"capability_non_primary_group_count":37},{"model":"DeepSeek V3","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2024-12-26","param_count":671.0,"active_param_count":37.0,"training_compute_flop":3.3e+24,"num_benchmarks":300,"benchmark_group_count":266,"benchmark_coverage_score":209.0,"source_family_count":81,"source_group_family_count":81,"source_effective_breadth":12.8,"source_primary":"benchmarklist.com","source_primary_share":0.1952,"source_primary_group_share":0.2143,"capability_breadth":7.31,"capability_primary":"knowledge_general","capability_primary_share":0.2192,"capability_primary_group_count":57,"capability_non_primary_group_count":209},{"model":"Qwen 3 0.6B (Non-reasoning)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":0.6,"active_param_count":null,"training_compute_flop":1.296e+23,"num_benchmarks":300,"benchmark_group_count":141,"benchmark_coverage_score":38.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.37,"source_primary":"euroeval.com","source_primary_share":0.8696,"source_primary_group_share":0.844,"capability_breadth":2.12,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6622,"capability_primary_group_count":103,"capability_non_primary_group_count":38},{"model":"Gemini 2.5 Flash Lite (Thinking)","provider":"Google","country":"US","openness":"api","release_date":"2025-09-25","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":299,"benchmark_group_count":139,"benchmark_coverage_score":36.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.33,"source_primary":"euroeval.com","source_primary_share":0.8725,"source_primary_group_share":0.8561,"capability_breadth":2.11,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6644,"capability_primary_group_count":103,"capability_non_primary_group_count":36},{"model":"Qwen 3 14B (Non-reasoning)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":14.0,"active_param_count":null,"training_compute_flop":3.197e+24,"num_benchmarks":299,"benchmark_group_count":140,"benchmark_coverage_score":37.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.35,"source_primary":"euroeval.com","source_primary_share":0.8725,"source_primary_group_share":0.85,"capability_breadth":2.11,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6644,"capability_primary_group_count":103,"capability_non_primary_group_count":37},{"model":"Bielik-11B-v2.3-Instruct","provider":"SpeakLeash","country":"PL","openness":"open","release_date":"2024-09-05","param_count":11.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":298,"benchmark_group_count":151,"benchmark_coverage_score":34.0,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":1.59,"source_primary":"euroeval.com","source_primary_share":0.8725,"source_primary_group_share":0.7881,"capability_breadth":1.79,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7282,"capability_primary_group_count":117,"capability_non_primary_group_count":34},{"model":"DeepSeek V4 Pro","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2026-04-23","param_count":1600.0,"active_param_count":49.0,"training_compute_flop":9.702e+24,"num_benchmarks":298,"benchmark_group_count":245,"benchmark_coverage_score":176.0,"source_family_count":89,"source_group_family_count":87,"source_effective_breadth":10.88,"source_primary":"benchmarklist.com","source_primary_share":0.2332,"source_primary_group_share":0.2653,"capability_breadth":6.4,"capability_primary":"code","capability_primary_share":0.2698,"capability_primary_group_count":64,"capability_non_primary_group_count":176},{"model":"Gemini 3 Flash (Preview) (Non-reasoning)","provider":"Google","country":"US","openness":"api","release_date":"2025-12-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":297,"benchmark_group_count":136,"benchmark_coverage_score":33.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"euroeval.com","source_primary_share":0.8754,"source_primary_group_share":0.875,"capability_breadth":2.08,"capability_primary":"knowledge_multilingual","capability_primary_share":0.67,"capability_primary_group_count":103,"capability_non_primary_group_count":33},{"model":"GPT-5.4 Mini (Non-reasoning)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-03-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":297,"benchmark_group_count":136,"benchmark_coverage_score":33.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.29,"source_primary":"euroeval.com","source_primary_share":0.8754,"source_primary_group_share":0.875,"capability_breadth":2.1,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6667,"capability_primary_group_count":103,"capability_non_primary_group_count":33},{"model":"Granite 4.0 H Small","provider":"IBM","country":"US","openness":"open","release_date":"2025-10-01","param_count":32.0,"active_param_count":9.0,"training_compute_flop":1.215e+24,"num_benchmarks":296,"benchmark_group_count":139,"benchmark_coverage_score":59.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.98,"source_primary":"euroeval.com","source_primary_share":0.7721,"source_primary_group_share":0.6835,"capability_breadth":2.55,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5714,"capability_primary_group_count":80,"capability_non_primary_group_count":59},{"model":"Gemini 2.5 Flash Lite (Non-reasoning)","provider":"Google","country":"US","openness":"api","release_date":"2025-06-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":296,"benchmark_group_count":137,"benchmark_coverage_score":34.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.3,"source_primary":"euroeval.com","source_primary_share":0.8784,"source_primary_group_share":0.8686,"capability_breadth":2.09,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6689,"capability_primary_group_count":103,"capability_non_primary_group_count":34},{"model":"GPT-OSS-20B (Medium)","provider":"OpenAI","country":"US","openness":"open","release_date":"2025-08-05","param_count":20.0,"active_param_count":3.61,"training_compute_flop":5.49e+23,"num_benchmarks":296,"benchmark_group_count":148,"benchmark_coverage_score":45.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":1.5,"source_primary":"euroeval.com","source_primary_share":0.8814,"source_primary_group_share":0.8041,"capability_breadth":2.01,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6712,"capability_primary_group_count":103,"capability_non_primary_group_count":45},{"model":"gemma-4-E2B-it","provider":"Google","country":"US","openness":"open","release_date":"2026-04-02","param_count":2.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":296,"benchmark_group_count":146,"benchmark_coverage_score":39.0,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":1.49,"source_primary":"euroeval.com","source_primary_share":0.8814,"source_primary_group_share":0.8151,"capability_breadth":2.01,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6847,"capability_primary_group_count":107,"capability_non_primary_group_count":39},{"model":"GLM-5","provider":"Zhipu","country":"CN","openness":"open","release_date":"2026-02-12","param_count":744.0,"active_param_count":40.0,"training_compute_flop":6.84e+24,"num_benchmarks":295,"benchmark_group_count":257,"benchmark_coverage_score":210.0,"source_family_count":74,"source_group_family_count":72,"source_effective_breadth":9.06,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.2297,"source_primary_group_share":0.249,"capability_breadth":9.17,"capability_primary":"reasoning_logical","capability_primary_share":0.1696,"capability_primary_group_count":47,"capability_non_primary_group_count":210},{"model":"GPT-5.4 Nano (Non-reasoning)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-03-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":295,"benchmark_group_count":136,"benchmark_coverage_score":33.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.29,"source_primary":"euroeval.com","source_primary_share":0.8814,"source_primary_group_share":0.875,"capability_breadth":2.07,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6712,"capability_primary_group_count":103,"capability_non_primary_group_count":33},{"model":"Qwen 3 8B (Non-reasoning)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":8.0,"active_param_count":null,"training_compute_flop":1.771e+24,"num_benchmarks":295,"benchmark_group_count":137,"benchmark_coverage_score":34.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.3,"source_primary":"euroeval.com","source_primary_share":0.8814,"source_primary_group_share":0.8686,"capability_breadth":2.07,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6712,"capability_primary_group_count":103,"capability_non_primary_group_count":34},{"model":"Apertus-70B-Instruct-2509","provider":"Apertus","country":"SG","openness":"open","release_date":"2025-09-01","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":294,"benchmark_group_count":135,"benchmark_coverage_score":32.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.27,"source_primary":"euroeval.com","source_primary_share":0.8874,"source_primary_group_share":0.8815,"capability_breadth":2.05,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6758,"capability_primary_group_count":103,"capability_non_primary_group_count":32},{"model":"Ministral-3-14B-Instruct-2512","provider":"Mistral","country":"FR","openness":"open","release_date":"2025-12-01","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":294,"benchmark_group_count":146,"benchmark_coverage_score":43.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.46,"source_primary":"euroeval.com","source_primary_share":0.8874,"source_primary_group_share":0.8151,"capability_breadth":1.98,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6758,"capability_primary_group_count":103,"capability_non_primary_group_count":43},{"model":"Ministral-3-14B-Reasoning-2512","provider":"Mistral","country":"FR","openness":"open","release_date":"2025-12-01","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":294,"benchmark_group_count":146,"benchmark_coverage_score":43.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":1.46,"source_primary":"euroeval.com","source_primary_share":0.8874,"source_primary_group_share":0.8151,"capability_breadth":1.98,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6758,"capability_primary_group_count":103,"capability_non_primary_group_count":43},{"model":"O3","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-04-16","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":293,"benchmark_group_count":237,"benchmark_coverage_score":191.0,"source_family_count":77,"source_group_family_count":77,"source_effective_breadth":19.31,"source_primary":"benchmarklist.com","source_primary_share":0.1429,"source_primary_group_share":0.1688,"capability_breadth":7.94,"capability_primary":"vision_language","capability_primary_share":0.1964,"capability_primary_group_count":46,"capability_non_primary_group_count":191},{"model":"Ministral-3-8B-Instruct-2512","provider":"Mistral","country":"FR","openness":"open","release_date":"2025-12-01","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":293,"benchmark_group_count":145,"benchmark_coverage_score":42.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.44,"source_primary":"euroeval.com","source_primary_share":0.8904,"source_primary_group_share":0.8207,"capability_breadth":1.97,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6781,"capability_primary_group_count":103,"capability_non_primary_group_count":42},{"model":"Apertus-8B-Instruct-2509","provider":"Apertus","country":"SG","openness":"open","release_date":"2025-09-01","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":292,"benchmark_group_count":129,"benchmark_coverage_score":36.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":1.67,"source_primary":"euroeval.com","source_primary_share":0.811,"source_primary_group_share":0.7597,"capability_breadth":2.14,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6564,"capability_primary_group_count":93,"capability_non_primary_group_count":36},{"model":"LFM2.5-1.2B Instruct","provider":"Liquid AI","country":"US","openness":"open","release_date":"2025-06-17","param_count":1.2,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":291,"benchmark_group_count":134,"benchmark_coverage_score":47.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":1.78,"source_primary":"euroeval.com","source_primary_share":0.8076,"source_primary_group_share":0.7313,"capability_breadth":2.38,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6186,"capability_primary_group_count":87,"capability_non_primary_group_count":47},{"model":"Granite 4.0 Micro","provider":"IBM","country":"US","openness":"open","release_date":"2025-10-01","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":291,"benchmark_group_count":138,"benchmark_coverage_score":38.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.38,"source_primary":"euroeval.com","source_primary_share":0.8655,"source_primary_group_share":0.8406,"capability_breadth":2.15,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6586,"capability_primary_group_count":100,"capability_non_primary_group_count":38},{"model":"Trinity-Mini","provider":"Arcee AI","country":"US","openness":"open","release_date":"2025-12-02","param_count":26.0,"active_param_count":3.0,"training_compute_flop":null,"num_benchmarks":291,"benchmark_group_count":143,"benchmark_coverage_score":40.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.4,"source_primary":"euroeval.com","source_primary_share":0.8966,"source_primary_group_share":0.8322,"capability_breadth":1.95,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6828,"capability_primary_group_count":103,"capability_non_primary_group_count":40},{"model":"Ministral-3-3B-Reasoning-2512","provider":"Mistral","country":"FR","openness":"open","release_date":"2025-12-01","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":290,"benchmark_group_count":143,"benchmark_coverage_score":40.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.4,"source_primary":"euroeval.com","source_primary_share":0.8966,"source_primary_group_share":0.8322,"capability_breadth":1.95,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6828,"capability_primary_group_count":103,"capability_non_primary_group_count":40},{"model":"GPT-5.4 Nano (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-03-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":287,"benchmark_group_count":144,"benchmark_coverage_score":41.0,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":1.45,"source_primary":"euroeval.com","source_primary_share":0.9059,"source_primary_group_share":0.8264,"capability_breadth":1.99,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6899,"capability_primary_group_count":103,"capability_non_primary_group_count":41},{"model":"AFM-4.5B","provider":"Other","country":"XX","openness":"open","release_date":"2025-09-01","param_count":4.5,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":286,"benchmark_group_count":138,"benchmark_coverage_score":47.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":1.86,"source_primary":"euroeval.com","source_primary_share":0.8252,"source_primary_group_share":0.7101,"capability_breadth":2.1,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6503,"capability_primary_group_count":91,"capability_non_primary_group_count":47},{"model":"Qwen 2.5 14B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-19","param_count":14.0,"active_param_count":null,"training_compute_flop":1.588e+24,"num_benchmarks":285,"benchmark_group_count":156,"benchmark_coverage_score":76.0,"source_family_count":29,"source_group_family_count":28,"source_effective_breadth":7.01,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.4346,"source_primary_group_share":0.2372,"capability_breadth":2.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4629,"capability_primary_group_count":80,"capability_non_primary_group_count":76},{"model":"salamandra-7B-instruct","provider":"BSC","country":"ES","openness":"open","release_date":"2024-09-30","param_count":7.8,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":285,"benchmark_group_count":142,"benchmark_coverage_score":30.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":1.41,"source_primary":"euroeval.com","source_primary_share":0.9123,"source_primary_group_share":0.838,"capability_breadth":1.77,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7298,"capability_primary_group_count":112,"capability_non_primary_group_count":30},{"model":"Magistral Small","provider":"Mistral","country":"FR","openness":"open","release_date":"2025-06-10","param_count":24.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":284,"benchmark_group_count":156,"benchmark_coverage_score":98.0,"source_family_count":21,"source_group_family_count":21,"source_effective_breadth":4.11,"source_primary":"euroeval.com","source_primary_share":0.5265,"source_primary_group_share":0.4487,"capability_breadth":3.76,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4028,"capability_primary_group_count":58,"capability_non_primary_group_count":98},{"model":"GPT-4o Mini (2024-07-18)","provider":"OpenAI","country":"US","openness":"api","release_date":"2024-07-18","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":284,"benchmark_group_count":234,"benchmark_coverage_score":174.0,"source_family_count":38,"source_group_family_count":38,"source_effective_breadth":6.73,"source_primary":"crfm.stanford.edu","source_primary_share":0.343,"source_primary_group_share":0.3462,"capability_breadth":5.86,"capability_primary":"knowledge_multilingual","capability_primary_share":0.2671,"capability_primary_group_count":60,"capability_non_primary_group_count":174},{"model":"Claude Opus 4.5","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-11-24","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":283,"benchmark_group_count":255,"benchmark_coverage_score":209.0,"source_family_count":87,"source_group_family_count":87,"source_effective_breadth":9.4,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.2426,"source_primary_group_share":0.2549,"capability_breadth":8.89,"capability_primary":"code","capability_primary_share":0.1765,"capability_primary_group_count":46,"capability_non_primary_group_count":209},{"model":"Gemini 3 Flash (Preview)","provider":"Google","country":"US","openness":"api","release_date":"2025-12-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":283,"benchmark_group_count":238,"benchmark_coverage_score":196.0,"source_family_count":56,"source_group_family_count":55,"source_effective_breadth":5.79,"source_primary":"benchmarklist.com","source_primary_share":0.3523,"source_primary_group_share":0.3866,"capability_breadth":8.14,"capability_primary":"code","capability_primary_share":0.178,"capability_primary_group_count":42,"capability_non_primary_group_count":196},{"model":"Gemma 2 27B","provider":"Google","country":"US","openness":"open","release_date":"2024-06-27","param_count":27.0,"active_param_count":null,"training_compute_flop":2.106e+24,"num_benchmarks":283,"benchmark_group_count":130,"benchmark_coverage_score":61.0,"source_family_count":25,"source_group_family_count":25,"source_effective_breadth":4.27,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.444,"source_primary_group_share":0.3846,"capability_breadth":2.75,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4549,"capability_primary_group_count":69,"capability_non_primary_group_count":61},{"model":"Granite 4.0 H 1B","provider":"IBM","country":"US","openness":"open","release_date":"2025-10-01","param_count":1.5,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":283,"benchmark_group_count":129,"benchmark_coverage_score":41.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.75,"source_primary":"euroeval.com","source_primary_share":0.805,"source_primary_group_share":0.7364,"capability_breadth":2.27,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6348,"capability_primary_group_count":88,"capability_non_primary_group_count":41},{"model":"Hunyuan A13B-Instruct","provider":"Tencent","country":"CN","openness":"open","release_date":"2025-06-25","param_count":80.0,"active_param_count":13.0,"training_compute_flop":null,"num_benchmarks":282,"benchmark_group_count":218,"benchmark_coverage_score":73.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/fair-forward/evals-for-every-language","source_primary_share":0.7158,"source_primary_group_share":0.6468,"capability_breadth":1.83,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7302,"capability_primary_group_count":145,"capability_non_primary_group_count":73},{"model":"GPT-4o ChatGPT","provider":"OpenAI","country":"US","openness":"api","release_date":"2024-05-13","param_count":200.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":282,"benchmark_group_count":166,"benchmark_coverage_score":57.0,"source_family_count":21,"source_group_family_count":21,"source_effective_breadth":3.52,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.4801,"source_primary_group_share":0.4398,"capability_breadth":1.81,"capability_primary":"vision_language","capability_primary_share":0.7329,"capability_primary_group_count":109,"capability_non_primary_group_count":57},{"model":"Qwen 1.5 14B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-02-05","param_count":14.0,"active_param_count":null,"training_compute_flop":3.36e+23,"num_benchmarks":282,"benchmark_group_count":139,"benchmark_coverage_score":49.0,"source_family_count":11,"source_group_family_count":10,"source_effective_breadth":4.1,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.4362,"source_primary_group_share":0.3165,"capability_breadth":2.46,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5213,"capability_primary_group_count":90,"capability_non_primary_group_count":49},{"model":"Qwen 3 235B A22B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":235.0,"active_param_count":22.0,"training_compute_flop":4.752e+24,"num_benchmarks":281,"benchmark_group_count":198,"benchmark_coverage_score":144.0,"source_family_count":61,"source_group_family_count":61,"source_effective_breadth":7.82,"source_primary":"euroeval.com","source_primary_share":0.365,"source_primary_group_share":0.2525,"capability_breadth":5.7,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3102,"capability_primary_group_count":54,"capability_non_primary_group_count":144},{"model":"GLM-4.7 Flash","provider":"Zhipu","country":"CN","openness":"open","release_date":"2026-01-19","param_count":30.0,"active_param_count":3.0,"training_compute_flop":null,"num_benchmarks":281,"benchmark_group_count":156,"benchmark_coverage_score":67.0,"source_family_count":22,"source_group_family_count":22,"source_effective_breadth":2.16,"source_primary":"euroeval.com","source_primary_share":0.7698,"source_primary_group_share":0.6731,"capability_breadth":2.61,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5899,"capability_primary_group_count":89,"capability_non_primary_group_count":67},{"model":"GPT-5 (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-08-07","param_count":null,"active_param_count":null,"training_compute_flop":6.6e+25,"num_benchmarks":280,"benchmark_group_count":178,"benchmark_coverage_score":133.0,"source_family_count":43,"source_group_family_count":43,"source_effective_breadth":8.83,"source_primary":"euroeval.com","source_primary_share":0.361,"source_primary_group_share":0.2809,"capability_breadth":5.86,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3069,"capability_primary_group_count":45,"capability_non_primary_group_count":133},{"model":"LFM2.5-1.2B (Thinking)","provider":"Liquid AI","country":"US","openness":"open","release_date":"2025-06-17","param_count":1.2,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":280,"benchmark_group_count":118,"benchmark_coverage_score":35.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":1.42,"source_primary":"euroeval.com","source_primary_share":0.8429,"source_primary_group_share":0.8305,"capability_breadth":2.3,"capability_primary":"knowledge_multilingual","capability_primary_share":0.625,"capability_primary_group_count":83,"capability_non_primary_group_count":35},{"model":"Claude 3 Opus","provider":"Anthropic","country":"US","openness":"api","release_date":"2024-03-04","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 3 family retired or superseded by newer Claude models.","num_benchmarks":278,"benchmark_group_count":198,"benchmark_coverage_score":81.0,"source_family_count":53,"source_group_family_count":52,"source_effective_breadth":3.72,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.5688,"source_primary_group_share":0.5051,"capability_breadth":2.34,"capability_primary":"vision_language","capability_primary_share":0.6341,"capability_primary_group_count":117,"capability_non_primary_group_count":81},{"model":"Llama 2 7B Chat","provider":"Meta","country":"US","openness":"open","release_date":"2023-07-18","param_count":7.0,"active_param_count":null,"training_compute_flop":8.4e+22,"num_benchmarks":278,"benchmark_group_count":176,"benchmark_coverage_score":96.0,"source_family_count":27,"source_group_family_count":27,"source_effective_breadth":7.16,"source_primary":"euroeval.com","source_primary_share":0.2826,"source_primary_group_share":0.2443,"capability_breadth":3.4,"capability_primary":"knowledge_multilingual","capability_primary_share":0.442,"capability_primary_group_count":80,"capability_non_primary_group_count":96},{"model":"EuroLLM-22B-Instruct-2512","provider":"EuroLLM","country":"XX","openness":"open","release_date":"2025-12-01","param_count":22.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":276,"benchmark_group_count":133,"benchmark_coverage_score":30.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.27,"source_primary":"euroeval.com","source_primary_share":0.913,"source_primary_group_share":0.8797,"capability_breadth":1.8,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7283,"capability_primary_group_count":103,"capability_non_primary_group_count":30},{"model":"Mistral Large 3","provider":"Mistral","country":"FR","openness":"api","release_date":"2025-12-02","param_count":675.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":274,"benchmark_group_count":212,"benchmark_coverage_score":165.0,"source_family_count":48,"source_group_family_count":48,"source_effective_breadth":7.7,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.2519,"source_primary_group_share":0.3066,"capability_breadth":8.67,"capability_primary":"reasoning_logical","capability_primary_share":0.187,"capability_primary_group_count":47,"capability_non_primary_group_count":165},{"model":"Llama 4 Scout Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2025-04-05","param_count":109.0,"active_param_count":17.0,"training_compute_flop":4.08e+24,"num_benchmarks":272,"benchmark_group_count":199,"benchmark_coverage_score":117.0,"source_family_count":33,"source_group_family_count":33,"source_effective_breadth":8.15,"source_primary":"euroeval.com","source_primary_share":0.3778,"source_primary_group_share":0.2513,"capability_breadth":4.18,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4333,"capability_primary_group_count":82,"capability_non_primary_group_count":117},{"model":"Gemini 2.5 Flash (Non-reasoning)","provider":"Google","country":"US","openness":"api","release_date":"2025-06-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":272,"benchmark_group_count":116,"benchmark_coverage_score":34.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.38,"source_primary":"euroeval.com","source_primary_share":0.864,"source_primary_group_share":0.8362,"capability_breadth":2.24,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6397,"capability_primary_group_count":82,"capability_non_primary_group_count":34},{"model":"Qwen 2.5 3B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-19","param_count":3.0,"active_param_count":null,"training_compute_flop":3.337e+23,"num_benchmarks":271,"benchmark_group_count":145,"benchmark_coverage_score":65.0,"source_family_count":24,"source_group_family_count":24,"source_effective_breadth":6.15,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.4556,"source_primary_group_share":0.2552,"capability_breadth":2.68,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4741,"capability_primary_group_count":80,"capability_non_primary_group_count":65},{"model":"Yi-1.5-9B","provider":"01.AI","country":"CN","openness":"open","release_date":"2024-05-13","param_count":9.0,"active_param_count":null,"training_compute_flop":1.907e+23,"num_benchmarks":270,"benchmark_group_count":119,"benchmark_coverage_score":53.0,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":3.56,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.4572,"source_primary_group_share":0.4202,"capability_breadth":2.64,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4572,"capability_primary_group_count":66,"capability_non_primary_group_count":53},{"model":"Grok 3","provider":"xAI","country":"US","openness":"api","release_date":"2025-02-17","param_count":null,"active_param_count":null,"training_compute_flop":3.5e+26,"num_benchmarks":269,"benchmark_group_count":186,"benchmark_coverage_score":144.0,"source_family_count":56,"source_group_family_count":56,"source_effective_breadth":10.09,"source_primary":"euroeval.com","source_primary_share":0.375,"source_primary_group_share":0.2634,"capability_breadth":6.47,"capability_primary":"knowledge_multilingual","capability_primary_share":0.2727,"capability_primary_group_count":42,"capability_non_primary_group_count":144},{"model":"Qwen 3.5 397B A17B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-02-16","param_count":397.0,"active_param_count":17.0,"training_compute_flop":null,"num_benchmarks":267,"benchmark_group_count":217,"benchmark_coverage_score":186.0,"source_family_count":55,"source_group_family_count":52,"source_effective_breadth":9.85,"source_primary":"benchmarklist.com","source_primary_share":0.2239,"source_primary_group_share":0.2673,"capability_breadth":9.07,"capability_primary":"knowledge_general","capability_primary_share":0.1467,"capability_primary_group_count":31,"capability_non_primary_group_count":186},{"model":"Ministral-3-3B-Instruct-2512","provider":"Mistral","country":"FR","openness":"open","release_date":"2025-12-01","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":267,"benchmark_group_count":133,"benchmark_coverage_score":39.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.41,"source_primary":"euroeval.com","source_primary_share":0.8914,"source_primary_group_share":0.8271,"capability_breadth":1.95,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6816,"capability_primary_group_count":94,"capability_non_primary_group_count":39},{"model":"NVIDIA Nemotron Nano 9B V2","provider":"NVIDIA","country":"US","openness":"open","release_date":"2025-08-18","param_count":9.0,"active_param_count":null,"training_compute_flop":1.53e+24,"num_benchmarks":266,"benchmark_group_count":160,"benchmark_coverage_score":87.0,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":2.33,"source_primary":"euroeval.com","source_primary_share":0.6988,"source_primary_group_share":0.5188,"capability_breadth":3.05,"capability_primary":"knowledge_multilingual","capability_primary_share":0.529,"capability_primary_group_count":73,"capability_non_primary_group_count":87},{"model":"MiniMax-M2.5","provider":"MiniMax","country":"CN","openness":"open","release_date":"2026-01-15","param_count":230.0,"active_param_count":10.0,"training_compute_flop":null,"num_benchmarks":264,"benchmark_group_count":212,"benchmark_coverage_score":173.0,"source_family_count":62,"source_group_family_count":59,"source_effective_breadth":16.77,"source_primary":"artificialanalysis.ai","source_primary_share":0.1853,"source_primary_group_share":0.1557,"capability_breadth":7.47,"capability_primary":"code","capability_primary_share":0.2046,"capability_primary_group_count":39,"capability_non_primary_group_count":173},{"model":"DeepSeek R1 Distill Llama 70B","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-01-20","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":263,"benchmark_group_count":171,"benchmark_coverage_score":82.0,"source_family_count":27,"source_group_family_count":27,"source_effective_breadth":6.43,"source_primary":"euroeval.com","source_primary_share":0.3831,"source_primary_group_share":0.2924,"capability_breadth":3.21,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5096,"capability_primary_group_count":89,"capability_non_primary_group_count":82},{"model":"Llama 2 70B Base","provider":"Meta","country":"US","openness":"open","release_date":"2023-07-18","param_count":70.0,"active_param_count":null,"training_compute_flop":8.1e+23,"num_benchmarks":263,"benchmark_group_count":146,"benchmark_coverage_score":69.0,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":4.6,"source_primary":"euroeval.com","source_primary_share":0.3817,"source_primary_group_share":0.3425,"capability_breadth":3.1,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4466,"capability_primary_group_count":77,"capability_non_primary_group_count":69},{"model":"Qwen 3 235B A22B 2507 Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-07-01","param_count":235.0,"active_param_count":22.0,"training_compute_flop":4.752e+24,"num_benchmarks":263,"benchmark_group_count":164,"benchmark_coverage_score":111.0,"source_family_count":33,"source_group_family_count":32,"source_effective_breadth":5.29,"source_primary":"euroeval.com","source_primary_share":0.5115,"source_primary_group_share":0.4024,"capability_breadth":4.77,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3779,"capability_primary_group_count":53,"capability_non_primary_group_count":111},{"model":"Trinity-Nano-Preview","provider":"Arcee AI","country":"US","openness":"open","release_date":"2026-01-27","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":263,"benchmark_group_count":119,"benchmark_coverage_score":36.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.41,"source_primary":"euroeval.com","source_primary_share":0.8973,"source_primary_group_share":0.8235,"capability_breadth":2.0,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6654,"capability_primary_group_count":83,"capability_non_primary_group_count":36},{"model":"GLM-4.7","provider":"Zhipu","country":"CN","openness":"open","release_date":"2025-12-22","param_count":355.0,"active_param_count":32.0,"training_compute_flop":4.42e+24,"num_benchmarks":260,"benchmark_group_count":230,"benchmark_coverage_score":187.0,"source_family_count":54,"source_group_family_count":53,"source_effective_breadth":8.75,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.2632,"source_primary_group_share":0.2783,"capability_breadth":8.24,"capability_primary":"reasoning_logical","capability_primary_share":0.1822,"capability_primary_group_count":43,"capability_non_primary_group_count":187},{"model":"GPT-5.4 Mini","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-03-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":256,"benchmark_group_count":229,"benchmark_coverage_score":193.0,"source_family_count":48,"source_group_family_count":48,"source_effective_breadth":6.39,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.2708,"source_primary_group_share":0.2795,"capability_breadth":9.0,"capability_primary":"knowledge_general","capability_primary_share":0.1506,"capability_primary_group_count":35,"capability_non_primary_group_count":193},{"model":"Qwen 2.5 7B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-19","param_count":7.0,"active_param_count":null,"training_compute_flop":8.219e+23,"num_benchmarks":255,"benchmark_group_count":135,"benchmark_coverage_score":65.0,"source_family_count":21,"source_group_family_count":21,"source_effective_breadth":5.46,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.4881,"source_primary_group_share":0.2741,"capability_breadth":2.67,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4484,"capability_primary_group_count":70,"capability_non_primary_group_count":65},{"model":"Llama 4 Scout","provider":"Meta","country":"US","openness":"open","release_date":"2025-04-05","param_count":109.0,"active_param_count":17.0,"training_compute_flop":4.08e+24,"num_benchmarks":249,"benchmark_group_count":191,"benchmark_coverage_score":163.0,"source_family_count":50,"source_group_family_count":50,"source_effective_breadth":6.28,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.2716,"source_primary_group_share":0.3403,"capability_breadth":9.01,"capability_primary":"knowledge_general","capability_primary_share":0.1687,"capability_primary_group_count":28,"capability_non_primary_group_count":163},{"model":"DeepSeek V3 (0324)","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-03-24","param_count":671.0,"active_param_count":37.0,"training_compute_flop":3.3e+24,"num_benchmarks":249,"benchmark_group_count":200,"benchmark_coverage_score":165.0,"source_family_count":55,"source_group_family_count":55,"source_effective_breadth":7.86,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.2686,"source_primary_group_share":0.32,"capability_breadth":7.84,"capability_primary":"knowledge_general","capability_primary_share":0.2107,"capability_primary_group_count":35,"capability_non_primary_group_count":165},{"model":"Llama 2 70B Chat (HF)","provider":"Meta","country":"US","openness":"open","release_date":"2023-07-18","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":248,"benchmark_group_count":140,"benchmark_coverage_score":63.0,"source_family_count":16,"source_group_family_count":16,"source_effective_breadth":4.36,"source_primary":"euroeval.com","source_primary_share":0.4049,"source_primary_group_share":0.3571,"capability_breadth":3.06,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4737,"capability_primary_group_count":77,"capability_non_primary_group_count":63},{"model":"Gemini 2.0 Flash (001)","provider":"Google","country":"US","openness":"api","release_date":"2025-02-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 2.0 family deprecated in favor of newer Gemini models.","num_benchmarks":246,"benchmark_group_count":164,"benchmark_coverage_score":116.0,"source_family_count":33,"source_group_family_count":33,"source_effective_breadth":7.65,"source_primary":"euroeval.com","source_primary_share":0.4167,"source_primary_group_share":0.3049,"capability_breadth":5.7,"capability_primary":"knowledge_multilingual","capability_primary_share":0.325,"capability_primary_group_count":48,"capability_non_primary_group_count":116},{"model":"Qwen3-8B-Base","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":246,"benchmark_group_count":63,"benchmark_coverage_score":33.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.33,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.5942,"source_primary_group_share":0.5714,"capability_breadth":2.26,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":30,"capability_non_primary_group_count":33},{"model":"DeepSeek R1 Distill Llama 8B","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-01-20","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":241,"benchmark_group_count":169,"benchmark_coverage_score":61.0,"source_family_count":18,"source_group_family_count":18,"source_effective_breadth":5.61,"source_primary":"euroeval.com","source_primary_share":0.4149,"source_primary_group_share":0.2959,"capability_breadth":2.33,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6266,"capability_primary_group_count":108,"capability_non_primary_group_count":61},{"model":"Qwen 2.5 32B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-19","param_count":32.0,"active_param_count":null,"training_compute_flop":3.51e+24,"num_benchmarks":241,"benchmark_group_count":109,"benchmark_coverage_score":54.0,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":4.06,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.5212,"source_primary_group_share":0.3394,"capability_breadth":2.94,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4068,"capability_primary_group_count":55,"capability_non_primary_group_count":54},{"model":"Qwen 2.5 0.5B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-19","param_count":0.5,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":240,"benchmark_group_count":119,"benchmark_coverage_score":49.0,"source_family_count":19,"source_group_family_count":19,"source_effective_breadth":4.35,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.5125,"source_primary_group_share":0.3109,"capability_breadth":2.54,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4917,"capability_primary_group_count":70,"capability_non_primary_group_count":49},{"model":"Gemini 1.0 Pro","provider":"Google","country":"US","openness":"api","release_date":"2024-02-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 1.x family retired or superseded by newer Gemini models.","num_benchmarks":239,"benchmark_group_count":167,"benchmark_coverage_score":54.0,"source_family_count":33,"source_group_family_count":33,"source_effective_breadth":2.66,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.6781,"source_primary_group_share":0.6048,"capability_breadth":1.82,"capability_primary":"vision_language","capability_primary_share":0.7339,"capability_primary_group_count":113,"capability_non_primary_group_count":54},{"model":"Kimi K2 (Thinking)","provider":"Moonshot","country":"CN","openness":"open","release_date":"2025-11-06","param_count":1000.0,"active_param_count":32.0,"training_compute_flop":4.2e+24,"num_benchmarks":236,"benchmark_group_count":180,"benchmark_coverage_score":144.0,"source_family_count":54,"source_group_family_count":53,"source_effective_breadth":17.61,"source_primary":"artificialanalysis.ai","source_primary_share":0.2227,"source_primary_group_share":0.1111,"capability_breadth":6.86,"capability_primary":"knowledge_general","capability_primary_share":0.2358,"capability_primary_group_count":36,"capability_non_primary_group_count":144},{"model":"GPT-5 Mini (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-08-07","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":236,"benchmark_group_count":143,"benchmark_coverage_score":98.0,"source_family_count":21,"source_group_family_count":21,"source_effective_breadth":5.71,"source_primary":"euroeval.com","source_primary_share":0.4237,"source_primary_group_share":0.3497,"capability_breadth":4.88,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3602,"capability_primary_group_count":45,"capability_non_primary_group_count":98},{"model":"DeepSeek R1 0528","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-05-28","param_count":671.0,"active_param_count":37.0,"training_compute_flop":4.02001e+24,"num_benchmarks":231,"benchmark_group_count":169,"benchmark_coverage_score":131.0,"source_family_count":59,"source_group_family_count":58,"source_effective_breadth":23.2,"source_primary":"artificialanalysis.ai","source_primary_share":0.2261,"source_primary_group_share":0.1243,"capability_breadth":6.4,"capability_primary":"code","capability_primary_share":0.2609,"capability_primary_group_count":38,"capability_non_primary_group_count":131},{"model":"MiniMax-M2.7","provider":"MiniMax","country":"CN","openness":"open","release_date":"2026-03-18","param_count":230.0,"active_param_count":10.0,"training_compute_flop":null,"num_benchmarks":230,"benchmark_group_count":187,"benchmark_coverage_score":129.0,"source_family_count":56,"source_group_family_count":53,"source_effective_breadth":6.65,"source_primary":"benchmarklist.com","source_primary_share":0.2889,"source_primary_group_share":0.3476,"capability_breadth":5.21,"capability_primary":"code","capability_primary_share":0.3125,"capability_primary_group_count":57,"capability_non_primary_group_count":129},{"model":"Llama 3 8B","provider":"Meta","country":"US","openness":"open","release_date":"2024-04-18","param_count":8.0,"active_param_count":null,"training_compute_flop":7.2e+23,"num_benchmarks":229,"benchmark_group_count":159,"benchmark_coverage_score":71.0,"source_family_count":34,"source_group_family_count":34,"source_effective_breadth":8.22,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.2203,"source_primary_group_share":0.2327,"capability_breadth":2.66,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5595,"capability_primary_group_count":88,"capability_non_primary_group_count":71},{"model":"Mistral-7B-v0.1","provider":"Mistral","country":"FR","openness":"open","release_date":"2023-09-27","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":225,"benchmark_group_count":137,"benchmark_coverage_score":89.0,"source_family_count":24,"source_group_family_count":24,"source_effective_breadth":9.2,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.2768,"source_primary_group_share":0.1971,"capability_breadth":3.71,"capability_primary":"knowledge_general","capability_primary_share":0.4554,"capability_primary_group_count":48,"capability_non_primary_group_count":89},{"model":"gemma-7B","provider":"Google","country":"US","openness":"open","release_date":"2024-02-21","param_count":7.0,"active_param_count":null,"training_compute_flop":3.07e+23,"num_benchmarks":225,"benchmark_group_count":139,"benchmark_coverage_score":62.0,"source_family_count":23,"source_group_family_count":23,"source_effective_breadth":5.61,"source_primary":"euroeval.com","source_primary_share":0.3795,"source_primary_group_share":0.3022,"capability_breadth":2.98,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5223,"capability_primary_group_count":77,"capability_non_primary_group_count":62},{"model":"Gemini 3 Pro","provider":"Google","country":"US","openness":"api","release_date":"2025-11-18","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":223,"benchmark_group_count":176,"benchmark_coverage_score":131.0,"source_family_count":72,"source_group_family_count":71,"source_effective_breadth":30.73,"source_primary":"fab-ai.org","source_primary_share":0.0972,"source_primary_group_share":0.0909,"capability_breadth":5.75,"capability_primary":"vision_language","capability_primary_share":0.2963,"capability_primary_group_count":45,"capability_non_primary_group_count":131},{"model":"Mistral Small 3","provider":"Mistral","country":"FR","openness":"open","release_date":"2025-01-30","param_count":24.0,"active_param_count":null,"training_compute_flop":1.152e+24,"num_benchmarks":223,"benchmark_group_count":143,"benchmark_coverage_score":80.0,"source_family_count":37,"source_group_family_count":37,"source_effective_breadth":6.49,"source_primary":"euroeval.com","source_primary_share":0.4566,"source_primary_group_share":0.3497,"capability_breadth":3.59,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4658,"capability_primary_group_count":63,"capability_non_primary_group_count":80},{"model":"O1","provider":"OpenAI","country":"US","openness":"api","release_date":"2024-12-05","param_count":200.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI o1 family deprecated or superseded by newer o-series models.","num_benchmarks":223,"benchmark_group_count":176,"benchmark_coverage_score":144.0,"source_family_count":60,"source_group_family_count":59,"source_effective_breadth":19.61,"source_primary":"artificialanalysis.ai","source_primary_share":0.1659,"source_primary_group_share":0.108,"capability_breadth":7.4,"capability_primary":"knowledge_general","capability_primary_share":0.2049,"capability_primary_group_count":32,"capability_non_primary_group_count":144},{"model":"Qwen 2 1.5B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-06-03","param_count":1.5,"active_param_count":null,"training_compute_flop":6.3e+22,"num_benchmarks":223,"benchmark_group_count":112,"benchmark_coverage_score":55.0,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":4.1,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.5516,"source_primary_group_share":0.3304,"capability_breadth":2.9,"capability_primary":"knowledge_multilingual","capability_primary_share":0.426,"capability_primary_group_count":57,"capability_non_primary_group_count":55},{"model":"Llama 2 7B Base","provider":"Meta","country":"US","openness":"open","release_date":"2023-07-18","param_count":7.0,"active_param_count":null,"training_compute_flop":8.4e+22,"num_benchmarks":220,"benchmark_group_count":117,"benchmark_coverage_score":56.0,"source_family_count":14,"source_group_family_count":14,"source_effective_breadth":4.99,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.2831,"source_primary_group_share":0.3162,"capability_breadth":3.08,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4292,"capability_primary_group_count":61,"capability_non_primary_group_count":56},{"model":"Qwen 2 72B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-06-10","param_count":72.0,"active_param_count":null,"training_compute_flop":3.02e+24,"num_benchmarks":219,"benchmark_group_count":170,"benchmark_coverage_score":104.0,"source_family_count":39,"source_group_family_count":38,"source_effective_breadth":9.9,"source_primary":"crfm.stanford.edu","source_primary_share":0.2385,"source_primary_group_share":0.2412,"capability_breadth":4.14,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4037,"capability_primary_group_count":66,"capability_non_primary_group_count":104},{"model":"Qwen 2.5 14B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-19","param_count":14.0,"active_param_count":null,"training_compute_flop":1.588e+24,"num_benchmarks":218,"benchmark_group_count":104,"benchmark_coverage_score":46.0,"source_family_count":14,"source_group_family_count":14,"source_effective_breadth":3.81,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.5642,"source_primary_group_share":0.3558,"capability_breadth":2.53,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4541,"capability_primary_group_count":58,"capability_non_primary_group_count":46},{"model":"GPT-4o (2024-05-13)","provider":"OpenAI","country":"US","openness":"api","release_date":"2024-05-13","param_count":200.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI GPT-4o snapshot deprecated in favor of newer GPT-4o/GPT-4.1/GPT-5 models.","num_benchmarks":215,"benchmark_group_count":172,"benchmark_coverage_score":118.0,"source_family_count":39,"source_group_family_count":39,"source_effective_breadth":8.84,"source_primary":"crfm.stanford.edu","source_primary_share":0.2871,"source_primary_group_share":0.2733,"capability_breadth":5.64,"capability_primary":"knowledge_general","capability_primary_share":0.2967,"capability_primary_group_count":54,"capability_non_primary_group_count":118},{"model":"O4 Mini (2025-04-16)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-04-16","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":214,"benchmark_group_count":151,"benchmark_coverage_score":101.0,"source_family_count":23,"source_group_family_count":23,"source_effective_breadth":4.97,"source_primary":"euroeval.com","source_primary_share":0.4808,"source_primary_group_share":0.3311,"capability_breadth":3.9,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3846,"capability_primary_group_count":50,"capability_non_primary_group_count":101},{"model":"Qwen 1.5 7B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-02-05","param_count":7.0,"active_param_count":null,"training_compute_flop":1.68e+23,"num_benchmarks":214,"benchmark_group_count":101,"benchmark_coverage_score":46.0,"source_family_count":15,"source_group_family_count":14,"source_effective_breadth":3.66,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.5748,"source_primary_group_share":0.3663,"capability_breadth":2.44,"capability_primary":"knowledge_general","capability_primary_share":0.472,"capability_primary_group_count":55,"capability_non_primary_group_count":46},{"model":"GPT-4 Turbo","provider":"OpenAI","country":"US","openness":"api","release_date":"2024-04-09","param_count":1760.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI GPT-4/GPT-4 Turbo family retired or superseded by newer API models.","num_benchmarks":213,"benchmark_group_count":175,"benchmark_coverage_score":141.0,"source_family_count":68,"source_group_family_count":67,"source_effective_breadth":24.4,"source_primary":"benchmarklist.com","source_primary_share":0.0913,"source_primary_group_share":0.1029,"capability_breadth":7.98,"capability_primary":"reasoning_logical","capability_primary_share":0.1971,"capability_primary_group_count":34,"capability_non_primary_group_count":141},{"model":"GLM-4.6","provider":"Zhipu","country":"CN","openness":"open","release_date":"2025-09-30","param_count":357.0,"active_param_count":32.0,"training_compute_flop":4.42e+24,"num_benchmarks":213,"benchmark_group_count":192,"benchmark_coverage_score":162.0,"source_family_count":44,"source_group_family_count":43,"source_effective_breadth":7.14,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.314,"source_primary_group_share":0.3333,"capability_breadth":9.42,"capability_primary":"creativity","capability_primary_share":0.1498,"capability_primary_group_count":30,"capability_non_primary_group_count":162},{"model":"DeepSeek-R1-Distill-Qwen-7B","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-01-20","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":212,"benchmark_group_count":138,"benchmark_coverage_score":49.0,"source_family_count":19,"source_group_family_count":19,"source_effective_breadth":4.54,"source_primary":"euroeval.com","source_primary_share":0.4717,"source_primary_group_share":0.3623,"capability_breadth":2.34,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6226,"capability_primary_group_count":89,"capability_non_primary_group_count":49},{"model":"Qwen 2.5 3B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-19","param_count":3.0,"active_param_count":null,"training_compute_flop":3.337e+23,"num_benchmarks":211,"benchmark_group_count":103,"benchmark_coverage_score":45.0,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":3.73,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.5829,"source_primary_group_share":0.3592,"capability_breadth":2.52,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4692,"capability_primary_group_count":58,"capability_non_primary_group_count":45},{"model":"MiniMax-M2.1","provider":"MiniMax","country":"CN","openness":"open","release_date":"2025-11-01","param_count":230.0,"active_param_count":10.0,"training_compute_flop":null,"num_benchmarks":209,"benchmark_group_count":150,"benchmark_coverage_score":106.0,"source_family_count":39,"source_group_family_count":38,"source_effective_breadth":12.47,"source_primary":"artificialanalysis.ai","source_primary_share":0.2488,"source_primary_group_share":0.14,"capability_breadth":5.58,"capability_primary":"knowledge_general","capability_primary_share":0.3317,"capability_primary_group_count":44,"capability_non_primary_group_count":106},{"model":"Qwen 2.5 1.5B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-19","param_count":1.5,"active_param_count":null,"training_compute_flop":1.663e+23,"num_benchmarks":209,"benchmark_group_count":101,"benchmark_coverage_score":42.0,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":3.6,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.5885,"source_primary_group_share":0.3663,"capability_breadth":2.45,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4785,"capability_primary_group_count":59,"capability_non_primary_group_count":42},{"model":"Qwen 2 0.5B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-06-03","param_count":0.5,"active_param_count":null,"training_compute_flop":3.6e+22,"num_benchmarks":209,"benchmark_group_count":105,"benchmark_coverage_score":52.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.63,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.5885,"source_primary_group_share":0.3524,"capability_breadth":2.83,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4306,"capability_primary_group_count":53,"capability_non_primary_group_count":52},{"model":"gemma-7B (IT)","provider":"Google","country":"US","openness":"open","release_date":"2024-02-21","param_count":7.0,"active_param_count":null,"training_compute_flop":3.07e+23,"num_benchmarks":207,"benchmark_group_count":126,"benchmark_coverage_score":51.0,"source_family_count":20,"source_group_family_count":20,"source_effective_breadth":4.74,"source_primary":"euroeval.com","source_primary_share":0.4146,"source_primary_group_share":0.3333,"capability_breadth":2.57,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5854,"capability_primary_group_count":75,"capability_non_primary_group_count":51},{"model":"Claude 3.7 Sonnet (20250219)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-02-19","param_count":null,"active_param_count":null,"training_compute_flop":3.35e+25,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 3 family retired or superseded by newer Claude models.","num_benchmarks":207,"benchmark_group_count":187,"benchmark_coverage_score":141.0,"source_family_count":37,"source_group_family_count":37,"source_effective_breadth":6.29,"source_primary":"crfm.stanford.edu","source_primary_share":0.3687,"source_primary_group_share":0.369,"capability_breadth":6.55,"capability_primary":"knowledge_general","capability_primary_share":0.2424,"capability_primary_group_count":46,"capability_non_primary_group_count":141},{"model":"Qwen 3.5 27B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-02-24","param_count":27.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":207,"benchmark_group_count":161,"benchmark_coverage_score":109.0,"source_family_count":37,"source_group_family_count":35,"source_effective_breadth":5.97,"source_primary":"llm-stats.com","source_primary_share":0.2759,"source_primary_group_share":0.3106,"capability_breadth":5.84,"capability_primary":"vision_language","capability_primary_share":0.3153,"capability_primary_group_count":52,"capability_non_primary_group_count":109},{"model":"Qwen 3 235B A22B (Thinking)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":235.0,"active_param_count":22.0,"training_compute_flop":4.752e+24,"num_benchmarks":206,"benchmark_group_count":147,"benchmark_coverage_score":95.0,"source_family_count":41,"source_group_family_count":40,"source_effective_breadth":10.54,"source_primary":"medmarks.ai","source_primary_share":0.2129,"source_primary_group_share":0.2245,"capability_breadth":4.7,"capability_primary":"knowledge_general","capability_primary_share":0.401,"capability_primary_group_count":52,"capability_non_primary_group_count":95},{"model":"Llama 3.1 405B Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-23","param_count":405.0,"active_param_count":null,"training_compute_flop":3.8e+25,"num_benchmarks":205,"benchmark_group_count":177,"benchmark_coverage_score":131.0,"source_family_count":40,"source_group_family_count":40,"source_effective_breadth":9.39,"source_primary":"crfm.stanford.edu","source_primary_share":0.2426,"source_primary_group_share":0.2542,"capability_breadth":6.63,"capability_primary":"knowledge_multilingual","capability_primary_share":0.2426,"capability_primary_group_count":46,"capability_non_primary_group_count":131},{"model":"Qwen 1.5 4B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-02-05","param_count":4.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":205,"benchmark_group_count":99,"benchmark_coverage_score":40.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.41,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.6,"source_primary_group_share":0.3737,"capability_breadth":2.44,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4634,"capability_primary_group_count":59,"capability_non_primary_group_count":40},{"model":"Gemini 3.5 Flash","provider":"Google","country":"US","openness":"api","release_date":"2026-05-19","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":204,"benchmark_group_count":175,"benchmark_coverage_score":142.0,"source_family_count":77,"source_group_family_count":77,"source_effective_breadth":19.75,"source_primary":"vals.ai","source_primary_share":0.135,"source_primary_group_share":0.1429,"capability_breadth":7.64,"capability_primary":"code","capability_primary_share":0.1709,"capability_primary_group_count":32,"capability_non_primary_group_count":142},{"model":"Qwen 2.5 72B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-19","param_count":72.0,"active_param_count":null,"training_compute_flop":7.8e+24,"num_benchmarks":204,"benchmark_group_count":85,"benchmark_coverage_score":56.0,"source_family_count":25,"source_group_family_count":25,"source_effective_breadth":4.83,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.6089,"source_primary_group_share":0.4235,"capability_breadth":3.22,"capability_primary":"knowledge_general","capability_primary_share":0.4802,"capability_primary_group_count":29,"capability_non_primary_group_count":56},{"model":"Falcon3-7B-Base","provider":"TII","country":"UAE","openness":"open","release_date":"2024-11-21","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":204,"benchmark_group_count":93,"benchmark_coverage_score":9.43,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.99,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.6029,"source_primary_group_share":0.3978,"capability_breadth":2.36,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5098,"capability_primary_group_count":64,"capability_non_primary_group_count":29},{"model":"Qwen 1.5 1.8B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-02-05","param_count":1.8,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":203,"benchmark_group_count":97,"benchmark_coverage_score":39.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":3.3,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.6059,"source_primary_group_share":0.3814,"capability_breadth":2.44,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4631,"capability_primary_group_count":58,"capability_non_primary_group_count":39},{"model":"Qwen 1.5 32B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-04-02","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":202,"benchmark_group_count":97,"benchmark_coverage_score":42.0,"source_family_count":12,"source_group_family_count":11,"source_effective_breadth":3.38,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.6089,"source_primary_group_share":0.3814,"capability_breadth":2.45,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4554,"capability_primary_group_count":55,"capability_non_primary_group_count":42},{"model":"GPT-4.5","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-02-27","param_count":null,"active_param_count":null,"training_compute_flop":3.8e+26,"lifecycle":"deprecated","lifecycle_reason":"OpenAI GPT-4.5 preview model retired or superseded by newer API models.","num_benchmarks":201,"benchmark_group_count":139,"benchmark_coverage_score":47.0,"source_family_count":24,"source_group_family_count":24,"source_effective_breadth":3.03,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.6617,"source_primary_group_share":0.554,"capability_breadth":1.75,"capability_primary":"vision_language","capability_primary_share":0.7512,"capability_primary_group_count":92,"capability_non_primary_group_count":47},{"model":"Mistral Nemo Instruct (2407)","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-07-18","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":201,"benchmark_group_count":146,"benchmark_coverage_score":64.0,"source_family_count":31,"source_group_family_count":31,"source_effective_breadth":8.69,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.2551,"source_primary_group_share":0.2534,"capability_breadth":2.62,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5816,"capability_primary_group_count":82,"capability_non_primary_group_count":64},{"model":"Kimi K2 0905","provider":"Moonshot","country":"CN","openness":"open","release_date":"2025-09-05","param_count":1000.0,"active_param_count":32.0,"training_compute_flop":null,"num_benchmarks":201,"benchmark_group_count":157,"benchmark_coverage_score":131.0,"source_family_count":42,"source_group_family_count":42,"source_effective_breadth":5.17,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.3351,"source_primary_group_share":0.4076,"capability_breadth":7.83,"capability_primary":"code","capability_primary_share":0.1856,"capability_primary_group_count":26,"capability_non_primary_group_count":131},{"model":"GPT-4o (2024-08-06)","provider":"OpenAI","country":"US","openness":"api","release_date":"2024-08-06","param_count":200.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":200,"benchmark_group_count":169,"benchmark_coverage_score":142.0,"source_family_count":50,"source_group_family_count":50,"source_effective_breadth":18.23,"source_primary":"artificialanalysis.ai","source_primary_share":0.1717,"source_primary_group_share":0.1065,"capability_breadth":8.0,"capability_primary":"code","capability_primary_share":0.1667,"capability_primary_group_count":27,"capability_non_primary_group_count":142},{"model":"DeepSeek R1 Distill Qwen 1.5B","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-01-20","param_count":1.5,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":199,"benchmark_group_count":130,"benchmark_coverage_score":48.0,"source_family_count":15,"source_group_family_count":15,"source_effective_breadth":4.1,"source_primary":"euroeval.com","source_primary_share":0.5025,"source_primary_group_share":0.3846,"capability_breadth":2.34,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6231,"capability_primary_group_count":82,"capability_non_primary_group_count":48},{"model":"O3 Mini (2025-01-31)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-01-31","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":199,"benchmark_group_count":135,"benchmark_coverage_score":87.0,"source_family_count":20,"source_group_family_count":20,"source_effective_breadth":4.47,"source_primary":"euroeval.com","source_primary_share":0.5208,"source_primary_group_share":0.3704,"capability_breadth":3.5,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4167,"capability_primary_group_count":48,"capability_non_primary_group_count":87},{"model":"Qwen 2.5 0.5B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-19","param_count":0.5,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":199,"benchmark_group_count":91,"benchmark_coverage_score":36.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":2.99,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.6181,"source_primary_group_share":0.4066,"capability_breadth":2.39,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4824,"capability_primary_group_count":55,"capability_non_primary_group_count":36},{"model":"Claude 3.5 Haiku (20241022)","provider":"Anthropic","country":"US","openness":"api","release_date":"2024-10-22","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 3 family retired or superseded by newer Claude models.","num_benchmarks":196,"benchmark_group_count":138,"benchmark_coverage_score":71.0,"source_family_count":25,"source_group_family_count":25,"source_effective_breadth":5.22,"source_primary":"euroeval.com","source_primary_share":0.4352,"source_primary_group_share":0.2971,"capability_breadth":3.17,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5181,"capability_primary_group_count":67,"capability_non_primary_group_count":71},{"model":"GPT-5 Nano (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-08-07","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":195,"benchmark_group_count":109,"benchmark_coverage_score":64.0,"source_family_count":16,"source_group_family_count":16,"source_effective_breadth":3.85,"source_primary":"euroeval.com","source_primary_share":0.5128,"source_primary_group_share":0.4587,"capability_breadth":3.91,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4359,"capability_primary_group_count":45,"capability_non_primary_group_count":64},{"model":"Llama 3.1 Nemotron 70B Instruct","provider":"NVIDIA","country":"US","openness":"open","release_date":"2024-10-15","param_count":70.0,"active_param_count":null,"training_compute_flop":7.929e+24,"num_benchmarks":194,"benchmark_group_count":125,"benchmark_coverage_score":64.0,"source_family_count":21,"source_group_family_count":21,"source_effective_breadth":6.25,"source_primary":"euroeval.com","source_primary_share":0.2577,"source_primary_group_share":0.296,"capability_breadth":3.6,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4691,"capability_primary_group_count":61,"capability_non_primary_group_count":64},{"model":"Nemotron Cascade 2 30B A3B","provider":"NVIDIA","country":"US","openness":"open","release_date":"2026-03-19","param_count":30.0,"active_param_count":3.0,"training_compute_flop":null,"num_benchmarks":191,"benchmark_group_count":97,"benchmark_coverage_score":9.32,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.39,"source_primary":"euroeval.com","source_primary_share":0.8168,"source_primary_group_share":0.8351,"capability_breadth":2.33,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6283,"capability_primary_group_count":69,"capability_non_primary_group_count":28},{"model":"GPT-5 (Minimal)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-08-07","param_count":null,"active_param_count":null,"training_compute_flop":6.6e+25,"num_benchmarks":190,"benchmark_group_count":98,"benchmark_coverage_score":53.0,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":3.15,"source_primary":"euroeval.com","source_primary_share":0.5291,"source_primary_group_share":0.5102,"capability_breadth":3.94,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4497,"capability_primary_group_count":45,"capability_non_primary_group_count":53},{"model":"Qwen 3.7 Max","provider":"Alibaba","country":"CN","openness":"api","release_date":"2026-05-20","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":189,"benchmark_group_count":142,"benchmark_coverage_score":94.0,"source_family_count":42,"source_group_family_count":41,"source_effective_breadth":8.68,"source_primary":"qwen.ai","source_primary_share":0.2099,"source_primary_group_share":0.2676,"capability_breadth":5.19,"capability_primary":"code","capability_primary_share":0.3184,"capability_primary_group_count":46,"capability_non_primary_group_count":94},{"model":"GPT-5 (2025-08-07)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-08-07","param_count":null,"active_param_count":null,"training_compute_flop":6.6e+25,"num_benchmarks":185,"benchmark_group_count":119,"benchmark_coverage_score":41.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.97,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.7348,"source_primary_group_share":0.6471,"capability_breadth":1.63,"capability_primary":"vision_language","capability_primary_share":0.7569,"capability_primary_group_count":78,"capability_non_primary_group_count":41},{"model":"GPT-5 Mini (2025-08-07)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-08-07","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":185,"benchmark_group_count":119,"benchmark_coverage_score":41.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.97,"source_primary":"huggingface.co/spaces/opencompass/open_vlm_leaderboard","source_primary_share":0.7348,"source_primary_group_share":0.6471,"capability_breadth":1.63,"capability_primary":"vision_language","capability_primary_share":0.7569,"capability_primary_group_count":78,"capability_non_primary_group_count":41},{"model":"Qwen 1.5 14B Chat","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-02-05","param_count":14.0,"active_param_count":null,"training_compute_flop":3.36e+23,"num_benchmarks":185,"benchmark_group_count":123,"benchmark_coverage_score":42.0,"source_family_count":17,"source_group_family_count":17,"source_effective_breadth":4.27,"source_primary":"euroeval.com","source_primary_share":0.4595,"source_primary_group_share":0.3577,"capability_breadth":2.24,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6432,"capability_primary_group_count":81,"capability_non_primary_group_count":42},{"model":"MamayLM-Gemma-3-12B-IT-v1.0","provider":"Other","country":"XX","openness":"open","release_date":"2025-05-01","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":184,"benchmark_group_count":45,"benchmark_coverage_score":8.96,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.95,"source_primary":"huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","source_primary_share":0.5241,"source_primary_group_share":0.5778,"capability_breadth":2.24,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5793,"capability_primary_group_count":32,"capability_non_primary_group_count":13},{"model":"lapa-v0.1.2-instruct","provider":"Other","country":"XX","openness":"open","release_date":"2025-10-18","param_count":12.2,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":184,"benchmark_group_count":45,"benchmark_coverage_score":8.96,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.95,"source_primary":"huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","source_primary_share":0.5241,"source_primary_group_share":0.5778,"capability_breadth":2.24,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5793,"capability_primary_group_count":32,"capability_non_primary_group_count":13},{"model":"lapa-12B-pt","provider":"Other","country":"XX","openness":"open","release_date":"2025-09-23","param_count":12.2,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":184,"benchmark_group_count":45,"benchmark_coverage_score":8.96,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.95,"source_primary":"huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","source_primary_share":0.5241,"source_primary_group_share":0.5778,"capability_breadth":2.24,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5793,"capability_primary_group_count":32,"capability_non_primary_group_count":13},{"model":"GPT-5.2 (Medium)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-12-11","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":183,"benchmark_group_count":123,"benchmark_coverage_score":85.0,"source_family_count":23,"source_group_family_count":23,"source_effective_breadth":6.9,"source_primary":"artificialanalysis.ai","source_primary_share":0.2905,"source_primary_group_share":0.2683,"capability_breadth":5.15,"capability_primary":"knowledge_general","capability_primary_share":0.3687,"capability_primary_group_count":38,"capability_non_primary_group_count":85},{"model":"Qwen2.5-32B-Instruct-abliterated-v2","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-10-11","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":183,"benchmark_group_count":82,"benchmark_coverage_score":31.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.48,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.6758,"source_primary_group_share":0.4512,"capability_breadth":2.39,"capability_primary":"knowledge_multilingual","capability_primary_share":0.467,"capability_primary_group_count":51,"capability_non_primary_group_count":31},{"model":"O4 Mini","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-04-16","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":182,"benchmark_group_count":153,"benchmark_coverage_score":120.0,"source_family_count":55,"source_group_family_count":55,"source_effective_breadth":11.22,"source_primary":"benchmarklist.com","source_primary_share":0.2286,"source_primary_group_share":0.2614,"capability_breadth":7.11,"capability_primary":"knowledge_general","capability_primary_share":0.2286,"capability_primary_group_count":33,"capability_non_primary_group_count":120},{"model":"GPT-4o (2024-11-20)","provider":"OpenAI","country":"US","openness":"api","release_date":"2024-11-20","param_count":200.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":181,"benchmark_group_count":150,"benchmark_coverage_score":124.0,"source_family_count":36,"source_group_family_count":36,"source_effective_breadth":10.3,"source_primary":"artificialanalysis.ai","source_primary_share":0.1989,"source_primary_group_share":0.2267,"capability_breadth":8.05,"capability_primary":"code","capability_primary_share":0.1818,"capability_primary_group_count":26,"capability_non_primary_group_count":124},{"model":"lambda-qwen2.5-14B-dpo-test","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-20","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":180,"benchmark_group_count":80,"benchmark_coverage_score":9.45,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.37,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.6833,"source_primary_group_share":0.4625,"capability_breadth":2.36,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4778,"capability_primary_group_count":52,"capability_non_primary_group_count":28},{"model":"Qwen 3 30B A3B (Non-reasoning)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":30.0,"active_param_count":3.0,"training_compute_flop":6.48e+23,"num_benchmarks":179,"benchmark_group_count":89,"benchmark_coverage_score":35.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.64,"source_primary":"euroeval.com","source_primary_share":0.7809,"source_primary_group_share":0.7528,"capability_breadth":2.68,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5674,"capability_primary_group_count":54,"capability_non_primary_group_count":35},{"model":"Qwen 3.5 122B A10B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-02-24","param_count":122.0,"active_param_count":10.0,"training_compute_flop":null,"num_benchmarks":179,"benchmark_group_count":141,"benchmark_coverage_score":97.0,"source_family_count":28,"source_group_family_count":27,"source_effective_breadth":7.09,"source_primary":"llm-stats.com","source_primary_share":0.3257,"source_primary_group_share":0.2837,"capability_breadth":6.32,"capability_primary":"vision_language","capability_primary_share":0.3143,"capability_primary_group_count":44,"capability_non_primary_group_count":97},{"model":"Awqward2.5-32B-Instruct","provider":"Other","country":"XX","openness":"open","release_date":"2024-12-18","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":179,"benchmark_group_count":79,"benchmark_coverage_score":9.46,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.31,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.6872,"source_primary_group_share":0.4684,"capability_breadth":2.37,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4749,"capability_primary_group_count":51,"capability_non_primary_group_count":28},{"model":"tempmotacilla-cinerea-0308","provider":"Other","country":"XX","openness":"open","release_date":"2025-03-08","param_count":14.8,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":179,"benchmark_group_count":79,"benchmark_coverage_score":9.46,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.31,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.6872,"source_primary_group_share":0.4684,"capability_breadth":2.37,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4749,"capability_primary_group_count":51,"capability_non_primary_group_count":28},{"model":"Qwentile2.5-32B-Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-12-19","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":179,"benchmark_group_count":79,"benchmark_coverage_score":9.46,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.31,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.6872,"source_primary_group_share":0.4684,"capability_breadth":2.37,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4749,"capability_primary_group_count":51,"capability_non_primary_group_count":28},{"model":"Saka-14B","provider":"Other","country":"XX","openness":"open","release_date":"2025-02-05","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":179,"benchmark_group_count":79,"benchmark_coverage_score":9.46,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.31,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.6872,"source_primary_group_share":0.4684,"capability_breadth":2.37,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4749,"capability_primary_group_count":51,"capability_non_primary_group_count":28},{"model":"oxyge1-33B","provider":"Other","country":"XX","openness":"open","release_date":"2024-12-08","param_count":33.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":179,"benchmark_group_count":79,"benchmark_coverage_score":9.46,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.31,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.6872,"source_primary_group_share":0.4684,"capability_breadth":2.37,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4749,"capability_primary_group_count":51,"capability_non_primary_group_count":28},{"model":"Dolphin3.0-R1-Mistral-24B","provider":"Cognitive Computations","country":"US","openness":"open","release_date":"2025-02-06","param_count":24.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":179,"benchmark_group_count":79,"benchmark_coverage_score":9.46,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.31,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.6872,"source_primary_group_share":0.4684,"capability_breadth":2.37,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4749,"capability_primary_group_count":51,"capability_non_primary_group_count":28},{"model":"DeepSeek-R1-ReDistill-Llama3-8B-v1.1","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-01-29","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":179,"benchmark_group_count":79,"benchmark_coverage_score":9.46,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.31,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.6872,"source_primary_group_share":0.4684,"capability_breadth":2.37,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4749,"capability_primary_group_count":51,"capability_non_primary_group_count":28},{"model":"Falcon3-10B-Base","provider":"TII","country":"UAE","openness":"open","release_date":"2024-12-03","param_count":10.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":179,"benchmark_group_count":79,"benchmark_coverage_score":9.46,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.31,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.6872,"source_primary_group_share":0.4684,"capability_breadth":2.37,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4749,"capability_primary_group_count":51,"capability_non_primary_group_count":28},{"model":"DeepSeek-R1-ReDistill-Qwen-7B-v1.1","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-01-27","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":179,"benchmark_group_count":79,"benchmark_coverage_score":9.46,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.31,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.6872,"source_primary_group_share":0.4684,"capability_breadth":2.37,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4749,"capability_primary_group_count":51,"capability_non_primary_group_count":28},{"model":"Qwen2.5-Math-7B-CFT","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-01-30","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":179,"benchmark_group_count":79,"benchmark_coverage_score":9.46,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.31,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.6872,"source_primary_group_share":0.4684,"capability_breadth":2.37,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4749,"capability_primary_group_count":51,"capability_non_primary_group_count":28},{"model":"Saka-1.5B","provider":"Other","country":"XX","openness":"open","release_date":"2025-02-07","param_count":1.5,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":179,"benchmark_group_count":79,"benchmark_coverage_score":9.46,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.31,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.6872,"source_primary_group_share":0.4684,"capability_breadth":2.37,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4749,"capability_primary_group_count":51,"capability_non_primary_group_count":28},{"model":"GPT-5.4 Nano","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-03-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":178,"benchmark_group_count":148,"benchmark_coverage_score":126.0,"source_family_count":29,"source_group_family_count":29,"source_effective_breadth":4.29,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.3824,"source_primary_group_share":0.4324,"capability_breadth":8.73,"capability_primary":"knowledge_general","capability_primary_share":0.1882,"capability_primary_group_count":22,"capability_non_primary_group_count":126},{"model":"Gemini 3 Flash","provider":"Google","country":"US","openness":"api","release_date":"2025-12-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":176,"benchmark_group_count":140,"benchmark_coverage_score":113.0,"source_family_count":53,"source_group_family_count":52,"source_effective_breadth":19.48,"source_primary":"fab-ai.org","source_primary_share":0.1228,"source_primary_group_share":0.1286,"capability_breadth":6.95,"capability_primary":"vision_language","capability_primary_share":0.2398,"capability_primary_group_count":27,"capability_non_primary_group_count":113},{"model":"Qwen 3.5 35B A3B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-02-24","param_count":35.0,"active_param_count":3.0,"training_compute_flop":null,"num_benchmarks":176,"benchmark_group_count":138,"benchmark_coverage_score":106.0,"source_family_count":32,"source_group_family_count":30,"source_effective_breadth":6.7,"source_primary":"llm-stats.com","source_primary_share":0.3275,"source_primary_group_share":0.2826,"capability_breadth":7.68,"capability_primary":"vision_language","capability_primary_share":0.2456,"capability_primary_group_count":32,"capability_non_primary_group_count":106},{"model":"Llama 2 13B Chat Base","provider":"Meta","country":"US","openness":"open","release_date":"2023-07-18","param_count":13.0,"active_param_count":null,"training_compute_flop":1.6e+23,"num_benchmarks":175,"benchmark_group_count":106,"benchmark_coverage_score":55.0,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":4.82,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.3563,"source_primary_group_share":0.3491,"capability_breadth":3.25,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3908,"capability_primary_group_count":51,"capability_non_primary_group_count":55},{"model":"NVIDIA Nemotron Nano 9B V2 (Non-reasoning)","provider":"NVIDIA","country":"US","openness":"open","release_date":"2025-08-18","param_count":9.0,"active_param_count":null,"training_compute_flop":1.53e+24,"num_benchmarks":173,"benchmark_group_count":84,"benchmark_coverage_score":31.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.48,"source_primary":"euroeval.com","source_primary_share":0.8035,"source_primary_group_share":0.7976,"capability_breadth":2.6,"capability_primary":"knowledge_multilingual","capability_primary_share":0.578,"capability_primary_group_count":53,"capability_non_primary_group_count":31},{"model":"Qwen2.5-32B-Instruct-CFT","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-01-30","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":173,"benchmark_group_count":73,"benchmark_coverage_score":9.07,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":2.0,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.711,"source_primary_group_share":0.5068,"capability_breadth":2.27,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4913,"capability_primary_group_count":51,"capability_non_primary_group_count":22},{"model":"EZO-Qwen2.5-32B-Instruct","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-21","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":173,"benchmark_group_count":73,"benchmark_coverage_score":9.07,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":2.0,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.711,"source_primary_group_share":0.5068,"capability_breadth":2.27,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4913,"capability_primary_group_count":51,"capability_non_primary_group_count":22},{"model":"gemma-2B","provider":"Google","country":"US","openness":"open","release_date":"2024-02-21","param_count":2.0,"active_param_count":null,"training_compute_flop":4.512e+22,"num_benchmarks":171,"benchmark_group_count":107,"benchmark_coverage_score":48.0,"source_family_count":15,"source_group_family_count":15,"source_effective_breadth":4.81,"source_primary":"euroeval.com","source_primary_share":0.3235,"source_primary_group_share":0.3458,"capability_breadth":3.32,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4824,"capability_primary_group_count":59,"capability_non_primary_group_count":48},{"model":"Qwen 3 1.7B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":1.7,"active_param_count":null,"training_compute_flop":3.672e+23,"num_benchmarks":171,"benchmark_group_count":90,"benchmark_coverage_score":30.0,"source_family_count":18,"source_group_family_count":18,"source_effective_breadth":1.88,"source_primary":"euroeval.com","source_primary_share":0.7661,"source_primary_group_share":0.7222,"capability_breadth":2.16,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6433,"capability_primary_group_count":60,"capability_non_primary_group_count":30},{"model":"Qwen 3 Max","provider":"Alibaba","country":"CN","openness":"api","release_date":"2025-09-24","param_count":700.0,"active_param_count":null,"training_compute_flop":1.512e+25,"num_benchmarks":170,"benchmark_group_count":126,"benchmark_coverage_score":102.0,"source_family_count":34,"source_group_family_count":34,"source_effective_breadth":12.03,"source_primary":"artificialanalysis.ai","source_primary_share":0.3036,"source_primary_group_share":0.1587,"capability_breadth":7.79,"capability_primary":"code","capability_primary_share":0.2024,"capability_primary_group_count":24,"capability_non_primary_group_count":102},{"model":"Grok 3 Mini","provider":"xAI","country":"US","openness":"api","release_date":"2025-02-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":170,"benchmark_group_count":103,"benchmark_coverage_score":62.0,"source_family_count":24,"source_group_family_count":24,"source_effective_breadth":4.03,"source_primary":"euroeval.com","source_primary_share":0.5858,"source_primary_group_share":0.4757,"capability_breadth":4.08,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4201,"capability_primary_group_count":41,"capability_non_primary_group_count":62},{"model":"O3 Mini","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-01-31","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":170,"benchmark_group_count":136,"benchmark_coverage_score":107.0,"source_family_count":56,"source_group_family_count":56,"source_effective_breadth":21.71,"source_primary":"benchmarklist.com","source_primary_share":0.0943,"source_primary_group_share":0.1103,"capability_breadth":7.66,"capability_primary":"knowledge_general","capability_primary_share":0.2264,"capability_primary_group_count":29,"capability_non_primary_group_count":107},{"model":"O3 (2025-04-16)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-04-16","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":169,"benchmark_group_count":114,"benchmark_coverage_score":71.0,"source_family_count":22,"source_group_family_count":22,"source_effective_breadth":4.35,"source_primary":"euroeval.com","source_primary_share":0.5988,"source_primary_group_share":0.4386,"capability_breadth":4.01,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4371,"capability_primary_group_count":43,"capability_non_primary_group_count":71},{"model":"Llama 3.1 405B","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-23","param_count":405.0,"active_param_count":null,"training_compute_flop":3.8e+25,"num_benchmarks":168,"benchmark_group_count":98,"benchmark_coverage_score":59.0,"source_family_count":29,"source_group_family_count":29,"source_effective_breadth":4.15,"source_primary":"euroeval.com","source_primary_share":0.561,"source_primary_group_share":0.4694,"capability_breadth":4.22,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4085,"capability_primary_group_count":39,"capability_non_primary_group_count":59},{"model":"GPT-OSS-120B (High)","provider":"OpenAI","country":"US","openness":"open","release_date":"2025-08-05","param_count":120.0,"active_param_count":5.13,"training_compute_flop":4.94e+24,"num_benchmarks":167,"benchmark_group_count":113,"benchmark_coverage_score":71.0,"source_family_count":22,"source_group_family_count":21,"source_effective_breadth":6.39,"source_primary":"artificialanalysis.ai","source_primary_share":0.3373,"source_primary_group_share":0.292,"capability_breadth":4.57,"capability_primary":"knowledge_general","capability_primary_share":0.3879,"capability_primary_group_count":41,"capability_non_primary_group_count":71},{"model":"SeaLLMs-v3-7B-Chat","provider":"Sea AI Lab","country":"SG","openness":"open","release_date":"2024-07-03","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":166,"benchmark_group_count":68,"benchmark_coverage_score":39.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.48,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.741,"source_primary_group_share":0.5294,"capability_breadth":2.53,"capability_primary":"knowledge_general","capability_primary_share":0.506,"capability_primary_group_count":29,"capability_non_primary_group_count":39},{"model":"Mistral Large 2 (Nov) Instruct (2411)","provider":"Mistral","country":"FR","openness":"api","release_date":"2024-11-18","param_count":123.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":165,"benchmark_group_count":126,"benchmark_coverage_score":103.0,"source_family_count":41,"source_group_family_count":41,"source_effective_breadth":15.81,"source_primary":"artificialanalysis.ai","source_primary_share":0.2188,"source_primary_group_share":0.1429,"capability_breadth":8.0,"capability_primary":"knowledge_general","capability_primary_share":0.2125,"capability_primary_group_count":23,"capability_non_primary_group_count":103},{"model":"MiniMax-M3","provider":"MiniMax","country":"CN","openness":"open","release_date":"2026-06-01","param_count":428.0,"active_param_count":23.0,"training_compute_flop":null,"num_benchmarks":165,"benchmark_group_count":128,"benchmark_coverage_score":95.0,"source_family_count":50,"source_group_family_count":50,"source_effective_breadth":13.15,"source_primary":"artificialanalysis.ai","source_primary_share":0.2236,"source_primary_group_share":0.1875,"capability_breadth":5.77,"capability_primary":"code","capability_primary_share":0.2687,"capability_primary_group_count":32,"capability_non_primary_group_count":95},{"model":"Llama 3 70B","provider":"Meta","country":"US","openness":"open","release_date":"2024-04-18","param_count":70.0,"active_param_count":null,"training_compute_flop":7.861e+24,"num_benchmarks":164,"benchmark_group_count":109,"benchmark_coverage_score":43.0,"source_family_count":28,"source_group_family_count":28,"source_effective_breadth":5.77,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.3145,"source_primary_group_share":0.3394,"capability_breadth":2.37,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6101,"capability_primary_group_count":66,"capability_non_primary_group_count":43},{"model":"Qwen 3.6 35B A3B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-04-16","param_count":35.0,"active_param_count":3.0,"training_compute_flop":null,"num_benchmarks":162,"benchmark_group_count":121,"benchmark_coverage_score":88.0,"source_family_count":35,"source_group_family_count":33,"source_effective_breadth":8.51,"source_primary":"llm-stats.com","source_primary_share":0.2375,"source_primary_group_share":0.2397,"capability_breadth":5.55,"capability_primary":"vision_language","capability_primary_share":0.2642,"capability_primary_group_count":32,"capability_non_primary_group_count":88},{"model":"GPT-4","provider":"OpenAI","country":"US","openness":"api","release_date":"2023-03-14","param_count":1760.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI GPT-4/GPT-4 Turbo family retired or superseded by newer API models.","num_benchmarks":160,"benchmark_group_count":109,"benchmark_coverage_score":84.0,"source_family_count":50,"source_group_family_count":49,"source_effective_breadth":17.76,"source_primary":"benchmarklist.com","source_primary_share":0.1301,"source_primary_group_share":0.1651,"capability_breadth":6.69,"capability_primary":"knowledge_general","capability_primary_share":0.2466,"capability_primary_group_count":25,"capability_non_primary_group_count":84},{"model":"DeepSeek V4 Flash","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2026-04-23","param_count":284.0,"active_param_count":13.0,"training_compute_flop":2.496e+24,"num_benchmarks":160,"benchmark_group_count":123,"benchmark_coverage_score":89.0,"source_family_count":51,"source_group_family_count":50,"source_effective_breadth":11.37,"source_primary":"benchmarklist.com","source_primary_share":0.1974,"source_primary_group_share":0.2439,"capability_breadth":6.12,"capability_primary":"code","capability_primary_share":0.2781,"capability_primary_group_count":33,"capability_non_primary_group_count":89},{"model":"Claude Fable 5","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-06-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":160,"benchmark_group_count":140,"benchmark_coverage_score":105.0,"source_family_count":70,"source_group_family_count":69,"source_effective_breadth":20.12,"source_primary":"vals.ai","source_primary_share":0.1529,"source_primary_group_share":0.1571,"capability_breadth":6.65,"capability_primary":"code","capability_primary_share":0.2258,"capability_primary_group_count":33,"capability_non_primary_group_count":105},{"model":"Qwen 2 72B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-06-10","param_count":72.0,"active_param_count":null,"training_compute_flop":3.02e+24,"num_benchmarks":157,"benchmark_group_count":61,"benchmark_coverage_score":38.0,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":2.66,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.7885,"source_primary_group_share":0.5902,"capability_breadth":2.37,"capability_primary":"knowledge_general","capability_primary_share":0.5705,"capability_primary_group_count":23,"capability_non_primary_group_count":38},{"model":"GPT-5 (Medium)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-08-07","param_count":null,"active_param_count":null,"training_compute_flop":6.6e+25,"num_benchmarks":156,"benchmark_group_count":103,"benchmark_coverage_score":81.0,"source_family_count":35,"source_group_family_count":35,"source_effective_breadth":11.06,"source_primary":"artificialanalysis.ai","source_primary_share":0.3464,"source_primary_group_share":0.2039,"capability_breadth":7.65,"capability_primary":"knowledge_general","capability_primary_share":0.1765,"capability_primary_group_count":22,"capability_non_primary_group_count":81},{"model":"Gemini 2.5 Flash (Preview 04-17)","provider":"Google","country":"US","openness":"api","release_date":"2025-04-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 2.5 preview snapshot retired or superseded by newer Gemini models.","num_benchmarks":156,"benchmark_group_count":109,"benchmark_coverage_score":70.0,"source_family_count":26,"source_group_family_count":26,"source_effective_breadth":4.84,"source_primary":"euroeval.com","source_primary_share":0.5935,"source_primary_group_share":0.422,"capability_breadth":4.2,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4323,"capability_primary_group_count":39,"capability_non_primary_group_count":70},{"model":"Mistral 7B Instruct (v0.3)","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-05-22","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":155,"benchmark_group_count":128,"benchmark_coverage_score":83.0,"source_family_count":36,"source_group_family_count":36,"source_effective_breadth":14.3,"source_primary":"crfm.stanford.edu","source_primary_share":0.1753,"source_primary_group_share":0.1328,"capability_breadth":4.44,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3896,"capability_primary_group_count":45,"capability_non_primary_group_count":83},{"model":"aya-23-8B","provider":"Cohere","country":"CA","openness":"open","release_date":"2024-05-23","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":155,"benchmark_group_count":87,"benchmark_coverage_score":7.69,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":3.72,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.3247,"source_primary_group_share":0.4253,"capability_breadth":1.92,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6883,"capability_primary_group_count":68,"capability_non_primary_group_count":19},{"model":"Yi-1.5-9B-32K","provider":"01.AI","country":"CN","openness":"open","release_date":"2024-05-13","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":155,"benchmark_group_count":58,"benchmark_coverage_score":36.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.38,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.7935,"source_primary_group_share":0.6207,"capability_breadth":2.31,"capability_primary":"knowledge_general","capability_primary_share":0.5935,"capability_primary_group_count":22,"capability_non_primary_group_count":36},{"model":"GPT-5 Mini (Minimal)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-08-07","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":153,"benchmark_group_count":79,"benchmark_coverage_score":37.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.19,"source_primary":"euroeval.com","source_primary_share":0.6536,"source_primary_group_share":0.6329,"capability_breadth":3.58,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4706,"capability_primary_group_count":42,"capability_non_primary_group_count":37},{"model":"Reka Flash 3","provider":"Reka","country":"US","openness":"open","release_date":"2025-03-10","param_count":21.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":152,"benchmark_group_count":84,"benchmark_coverage_score":43.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":2.45,"source_primary":"euroeval.com","source_primary_share":0.6623,"source_primary_group_share":0.5952,"capability_breadth":3.61,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4702,"capability_primary_group_count":41,"capability_non_primary_group_count":43},{"model":"GPT-5 Nano (Minimal)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-08-07","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":152,"benchmark_group_count":78,"benchmark_coverage_score":36.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.14,"source_primary":"euroeval.com","source_primary_share":0.6579,"source_primary_group_share":0.641,"capability_breadth":3.54,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4737,"capability_primary_group_count":42,"capability_non_primary_group_count":36},{"model":"SmolLM3-3B","provider":"HuggingFace","country":"FR","openness":"open","release_date":"2025-07-01","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":150,"benchmark_group_count":85,"benchmark_coverage_score":37.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.33,"source_primary":"euroeval.com","source_primary_share":0.6824,"source_primary_group_share":0.6,"capability_breadth":2.51,"capability_primary":"knowledge_multilingual","capability_primary_share":0.527,"capability_primary_group_count":48,"capability_non_primary_group_count":37},{"model":"SeaLLM-7B-v2.5","provider":"Sea AI Lab","country":"SG","openness":"open","release_date":"2024-04-10","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":150,"benchmark_group_count":54,"benchmark_coverage_score":33.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.1,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.82,"source_primary_group_share":0.6667,"capability_breadth":2.18,"capability_primary":"knowledge_general","capability_primary_share":0.6067,"capability_primary_group_count":21,"capability_non_primary_group_count":33},{"model":"Grok 4.1 Fast","provider":"xAI","country":"US","openness":"api","release_date":"2025-11-19","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":150,"benchmark_group_count":118,"benchmark_coverage_score":93.0,"source_family_count":40,"source_group_family_count":40,"source_effective_breadth":8.76,"source_primary":"benchmarklist.com","source_primary_share":0.234,"source_primary_group_share":0.2797,"capability_breadth":7.29,"capability_primary":"knowledge_general","capability_primary_share":0.2357,"capability_primary_group_count":24,"capability_non_primary_group_count":93},{"model":"DeepSeek R1 0528 Qwen3 8B","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-05-28","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":150,"benchmark_group_count":80,"benchmark_coverage_score":38.0,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":2.25,"source_primary":"euroeval.com","source_primary_share":0.6667,"source_primary_group_share":0.625,"capability_breadth":3.37,"capability_primary":"knowledge_multilingual","capability_primary_share":0.48,"capability_primary_group_count":42,"capability_non_primary_group_count":38},{"model":"Grok 4.3","provider":"xAI","country":"US","openness":"api","release_date":"2026-04-30","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":148,"benchmark_group_count":121,"benchmark_coverage_score":94.0,"source_family_count":48,"source_group_family_count":47,"source_effective_breadth":10.65,"source_primary":"vals.ai","source_primary_share":0.2113,"source_primary_group_share":0.2314,"capability_breadth":7.3,"capability_primary":"knowledge_general","capability_primary_share":0.2324,"capability_primary_group_count":27,"capability_non_primary_group_count":94},{"model":"Mistral 7B Instruct (v0.2)","provider":"Mistral","country":"FR","openness":"open","release_date":"2023-12-11","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":148,"benchmark_group_count":115,"benchmark_coverage_score":76.0,"source_family_count":34,"source_group_family_count":34,"source_effective_breadth":12.14,"source_primary":"euroeval.com","source_primary_share":0.2517,"source_primary_group_share":0.1739,"capability_breadth":4.18,"capability_primary":"knowledge_multilingual","capability_primary_share":0.415,"capability_primary_group_count":39,"capability_non_primary_group_count":76},{"model":"Phi-3.5-mini-instruct","provider":"Microsoft","country":"US","openness":"open","release_date":"2024-08-20","param_count":3.8,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":146,"benchmark_group_count":120,"benchmark_coverage_score":64.0,"source_family_count":23,"source_group_family_count":22,"source_effective_breadth":6.61,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.3425,"source_primary_group_share":0.3083,"capability_breadth":3.53,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4932,"capability_primary_group_count":56,"capability_non_primary_group_count":64},{"model":"Claude Opus 4.6 (Thinking)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-02-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":145,"benchmark_group_count":133,"benchmark_coverage_score":101.0,"source_family_count":20,"source_group_family_count":20,"source_effective_breadth":6.39,"source_primary":"qwen.ai","source_primary_share":0.2609,"source_primary_group_share":0.2707,"capability_breadth":5.59,"capability_primary":"reasoning_logical","capability_primary_share":0.2391,"capability_primary_group_count":32,"capability_non_primary_group_count":101},{"model":"Qwen 3 VL 8B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-07-01","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":145,"benchmark_group_count":104,"benchmark_coverage_score":53.0,"source_family_count":16,"source_group_family_count":16,"source_effective_breadth":6.56,"source_primary":"llm-stats.com","source_primary_share":0.2571,"source_primary_group_share":0.2212,"capability_breadth":4.09,"capability_primary":"vision_language","capability_primary_share":0.4429,"capability_primary_group_count":51,"capability_non_primary_group_count":53},{"model":"gemma-2B (IT)","provider":"Google","country":"US","openness":"open","release_date":"2024-02-21","param_count":2.0,"active_param_count":null,"training_compute_flop":4.512e+22,"num_benchmarks":144,"benchmark_group_count":98,"benchmark_coverage_score":41.0,"source_family_count":15,"source_group_family_count":15,"source_effective_breadth":4.26,"source_primary":"euroeval.com","source_primary_share":0.3636,"source_primary_group_share":0.3776,"capability_breadth":2.59,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5874,"capability_primary_group_count":57,"capability_non_primary_group_count":41},{"model":"MiMo-V2.5-Pro","provider":"Xiaomi","country":"CN","openness":"open","release_date":"2026-04-23","param_count":1020.0,"active_param_count":42.0,"training_compute_flop":6.804e+24,"num_benchmarks":144,"benchmark_group_count":106,"benchmark_coverage_score":70.0,"source_family_count":35,"source_group_family_count":34,"source_effective_breadth":9.91,"source_primary":"artificialanalysis.ai","source_primary_share":0.2571,"source_primary_group_share":0.1981,"capability_breadth":4.91,"capability_primary":"code","capability_primary_share":0.3333,"capability_primary_group_count":34,"capability_non_primary_group_count":70},{"model":"MiniMax-M2","provider":"MiniMax","country":"CN","openness":"open","release_date":"2025-10-23","param_count":230.0,"active_param_count":10.0,"training_compute_flop":null,"num_benchmarks":144,"benchmark_group_count":101,"benchmark_coverage_score":68.0,"source_family_count":33,"source_group_family_count":33,"source_effective_breadth":9.65,"source_primary":"artificialanalysis.ai","source_primary_share":0.3617,"source_primary_group_share":0.2079,"capability_breadth":4.64,"capability_primary":"knowledge_general","capability_primary_share":0.3617,"capability_primary_group_count":33,"capability_non_primary_group_count":68},{"model":"GLM-5.2","provider":"Zhipu","country":"CN","openness":"open","release_date":"2026-06-13","param_count":744.0,"active_param_count":40.0,"training_compute_flop":null,"num_benchmarks":144,"benchmark_group_count":132,"benchmark_coverage_score":91.0,"source_family_count":64,"source_group_family_count":63,"source_effective_breadth":19.71,"source_primary":"vals.ai","source_primary_share":0.1549,"source_primary_group_share":0.1515,"capability_breadth":5.98,"capability_primary":"code","capability_primary_share":0.2929,"capability_primary_group_count":39,"capability_non_primary_group_count":91},{"model":"GPT-5.5 (xHigh)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-04-23","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":140,"benchmark_group_count":109,"benchmark_coverage_score":78.0,"source_family_count":34,"source_group_family_count":33,"source_effective_breadth":11.01,"source_primary":"artificialanalysis.ai","source_primary_share":0.2701,"source_primary_group_share":0.1651,"capability_breadth":5.21,"capability_primary":"code","capability_primary_share":0.2868,"capability_primary_group_count":30,"capability_non_primary_group_count":78},{"model":"Gemma 2 9B","provider":"Google","country":"US","openness":"open","release_date":"2024-06-27","param_count":9.0,"active_param_count":null,"training_compute_flop":4.32e+23,"num_benchmarks":140,"benchmark_group_count":91,"benchmark_coverage_score":44.0,"source_family_count":20,"source_group_family_count":20,"source_effective_breadth":4.15,"source_primary":"euroeval.com","source_primary_share":0.6115,"source_primary_group_share":0.4615,"capability_breadth":2.71,"capability_primary":"knowledge_multilingual","capability_primary_share":0.554,"capability_primary_group_count":47,"capability_non_primary_group_count":44},{"model":"Step 3.5 Flash","provider":"StepFun","country":"CN","openness":"api","release_date":"2026-02-02","param_count":196.0,"active_param_count":11.0,"training_compute_flop":null,"num_benchmarks":140,"benchmark_group_count":104,"benchmark_coverage_score":74.0,"source_family_count":28,"source_group_family_count":26,"source_effective_breadth":9.64,"source_primary":"artificialanalysis.ai","source_primary_share":0.223,"source_primary_group_share":0.1923,"capability_breadth":5.4,"capability_primary":"math","capability_primary_share":0.2878,"capability_primary_group_count":30,"capability_non_primary_group_count":74},{"model":"Claude 3.5 Haiku","provider":"Anthropic","country":"US","openness":"api","release_date":"2024-11-04","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 3 family retired or superseded by newer Claude models.","num_benchmarks":138,"benchmark_group_count":97,"benchmark_coverage_score":78.0,"source_family_count":32,"source_group_family_count":32,"source_effective_breadth":11.08,"source_primary":"artificialanalysis.ai","source_primary_share":0.2836,"source_primary_group_share":0.1856,"capability_breadth":7.65,"capability_primary":"knowledge_general","capability_primary_share":0.2463,"capability_primary_group_count":19,"capability_non_primary_group_count":78},{"model":"Qwen 1.5 0.5B Chat","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-02-04","param_count":0.5,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":138,"benchmark_group_count":101,"benchmark_coverage_score":45.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":4.33,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.3623,"source_primary_group_share":0.3663,"capability_breadth":2.84,"capability_primary":"knowledge_multilingual","capability_primary_share":0.558,"capability_primary_group_count":56,"capability_non_primary_group_count":45},{"model":"free-evo-qwen72B-v0.8-re","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-02","param_count":72.3,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":137,"benchmark_group_count":49,"benchmark_coverage_score":30.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.74,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.8978,"source_primary_group_share":0.7347,"capability_breadth":2.25,"capability_primary":"knowledge_general","capability_primary_share":0.5912,"capability_primary_group_count":19,"capability_non_primary_group_count":30},{"model":"Yi-1.5-34B-32K","provider":"01.AI","country":"CN","openness":"open","release_date":"2024-05-15","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":137,"benchmark_group_count":49,"benchmark_coverage_score":30.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.74,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.8978,"source_primary_group_share":0.7347,"capability_breadth":2.25,"capability_primary":"knowledge_general","capability_primary_share":0.5912,"capability_primary_group_count":19,"capability_non_primary_group_count":30},{"model":"Qwen 3 VL 235B A22B (Thinking)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-09-22","param_count":235.0,"active_param_count":22.0,"training_compute_flop":null,"num_benchmarks":137,"benchmark_group_count":96,"benchmark_coverage_score":56.0,"source_family_count":17,"source_group_family_count":17,"source_effective_breadth":5.86,"source_primary":"llm-stats.com","source_primary_share":0.3256,"source_primary_group_share":0.3021,"capability_breadth":4.58,"capability_primary":"vision_language","capability_primary_share":0.4031,"capability_primary_group_count":40,"capability_non_primary_group_count":56},{"model":"Bielik-11B-v2","provider":"SpeakLeash","country":"PL","openness":"open","release_date":"2024-08-26","param_count":11.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":137,"benchmark_group_count":46,"benchmark_coverage_score":9.22,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":1.58,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.8978,"source_primary_group_share":0.7826,"capability_breadth":2.3,"capability_primary":"knowledge_general","capability_primary_share":0.5547,"capability_primary_group_count":20,"capability_non_primary_group_count":26},{"model":"DeepSeek V3.1 Terminus","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-09-22","param_count":671.0,"active_param_count":37.0,"training_compute_flop":3.594058e+24,"num_benchmarks":136,"benchmark_group_count":110,"benchmark_coverage_score":88.0,"source_family_count":18,"source_group_family_count":18,"source_effective_breadth":2.79,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.5,"source_primary_group_share":0.5818,"capability_breadth":8.08,"capability_primary":"code","capability_primary_share":0.1923,"capability_primary_group_count":22,"capability_non_primary_group_count":88},{"model":"Mistral Large 2 (Jul)","provider":"Mistral","country":"FR","openness":"api","release_date":"2024-07-24","param_count":123.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":135,"benchmark_group_count":124,"benchmark_coverage_score":100.0,"source_family_count":47,"source_group_family_count":47,"source_effective_breadth":17.59,"source_primary":"livebench.ai","source_primary_share":0.1278,"source_primary_group_share":0.1371,"capability_breadth":7.52,"capability_primary":"knowledge_multilingual","capability_primary_share":0.1955,"capability_primary_group_count":24,"capability_non_primary_group_count":100},{"model":"Mixtral 8x7B Instruct (v0.1)","provider":"Mistral","country":"FR","openness":"open","release_date":"2023-12-11","param_count":46.7,"active_param_count":12.9,"training_compute_flop":7.74e+23,"num_benchmarks":135,"benchmark_group_count":104,"benchmark_coverage_score":74.0,"source_family_count":32,"source_group_family_count":32,"source_effective_breadth":9.96,"source_primary":"euroeval.com","source_primary_share":0.3459,"source_primary_group_share":0.2212,"capability_breadth":4.68,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3609,"capability_primary_group_count":30,"capability_non_primary_group_count":74},{"model":"Claude Opus 4","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-05-22","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 4.0 family deprecated in favor of newer Claude 4.x models.","num_benchmarks":134,"benchmark_group_count":106,"benchmark_coverage_score":88.0,"source_family_count":47,"source_group_family_count":47,"source_effective_breadth":18.42,"source_primary":"fab-ai.org","source_primary_share":0.1628,"source_primary_group_share":0.1604,"capability_breadth":7.63,"capability_primary":"vision_language","capability_primary_share":0.186,"capability_primary_group_count":18,"capability_non_primary_group_count":88},{"model":"Gemini 2.5 Pro (Preview 03-25)","provider":"Google","country":"US","openness":"api","release_date":"2025-03-25","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 2.5 preview snapshot retired or superseded by newer Gemini models.","num_benchmarks":134,"benchmark_group_count":82,"benchmark_coverage_score":39.0,"source_family_count":25,"source_group_family_count":25,"source_effective_breadth":2.62,"source_primary":"euroeval.com","source_primary_share":0.7576,"source_primary_group_share":0.6098,"capability_breadth":2.94,"capability_primary":"knowledge_multilingual","capability_primary_share":0.553,"capability_primary_group_count":43,"capability_non_primary_group_count":39},{"model":"SmolLM2-1.7B-Instruct","provider":"HuggingFace","country":"FR","openness":"open","release_date":"2024-10-31","param_count":1.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":133,"benchmark_group_count":93,"benchmark_coverage_score":8.2,"source_family_count":14,"source_group_family_count":14,"source_effective_breadth":4.39,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.3788,"source_primary_group_share":0.3978,"capability_breadth":2.05,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6742,"capability_primary_group_count":65,"capability_non_primary_group_count":28},{"model":"Qwen 3.7 Plus","provider":"Alibaba","country":"CN","openness":"api","release_date":"2026-06-02","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":133,"benchmark_group_count":95,"benchmark_coverage_score":63.0,"source_family_count":23,"source_group_family_count":23,"source_effective_breadth":6.66,"source_primary":"artificialanalysis.ai","source_primary_share":0.2727,"source_primary_group_share":0.2526,"capability_breadth":5.72,"capability_primary":"vision_language","capability_primary_share":0.303,"capability_primary_group_count":32,"capability_non_primary_group_count":63},{"model":"SILMA-9B-Instruct-v1.0","provider":"SILMA AI","country":"US","openness":"open","release_date":"2024-08-17","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":132,"benchmark_group_count":45,"benchmark_coverage_score":9.06,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.52,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.9318,"source_primary_group_share":0.8,"capability_breadth":2.27,"capability_primary":"knowledge_general","capability_primary_share":0.5758,"capability_primary_group_count":19,"capability_non_primary_group_count":26},{"model":"Phi-4 (Reasoning)","provider":"Microsoft","country":"US","openness":"open","release_date":"2025-04-30","param_count":14.0,"active_param_count":null,"training_compute_flop":9.32e+23,"num_benchmarks":132,"benchmark_group_count":76,"benchmark_coverage_score":35.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":1.96,"source_primary":"euroeval.com","source_primary_share":0.7576,"source_primary_group_share":0.6579,"capability_breadth":2.53,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5379,"capability_primary_group_count":41,"capability_non_primary_group_count":35},{"model":"O4 Mini (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-04-16","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":130,"benchmark_group_count":105,"benchmark_coverage_score":76.0,"source_family_count":40,"source_group_family_count":40,"source_effective_breadth":14.12,"source_primary":"artificialanalysis.ai","source_primary_share":0.2769,"source_primary_group_share":0.1714,"capability_breadth":5.72,"capability_primary":"code","capability_primary_share":0.2923,"capability_primary_group_count":29,"capability_non_primary_group_count":76},{"model":"recoilme-gemma-2-9B-v0.4","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-18","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":130,"benchmark_group_count":43,"benchmark_coverage_score":8.98,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.39,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.9462,"source_primary_group_share":0.8372,"capability_breadth":2.25,"capability_primary":"knowledge_general","capability_primary_share":0.5846,"capability_primary_group_count":17,"capability_non_primary_group_count":26},{"model":"Qwen2.5-14B-Gutenberg-1e-Delta","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-20","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":130,"benchmark_group_count":43,"benchmark_coverage_score":8.98,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.39,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.9462,"source_primary_group_share":0.8372,"capability_breadth":2.25,"capability_primary":"knowledge_general","capability_primary_share":0.5846,"capability_primary_group_count":17,"capability_non_primary_group_count":26},{"model":"Qwen2.5-Lumen-14B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-20","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":130,"benchmark_group_count":43,"benchmark_coverage_score":8.98,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.39,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.9462,"source_primary_group_share":0.8372,"capability_breadth":2.25,"capability_primary":"knowledge_general","capability_primary_share":0.5846,"capability_primary_group_count":17,"capability_non_primary_group_count":26},{"model":"recoilme-gemma-2-9B-v0.2","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-18","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":130,"benchmark_group_count":43,"benchmark_coverage_score":8.98,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.39,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.9462,"source_primary_group_share":0.8372,"capability_breadth":2.25,"capability_primary":"knowledge_general","capability_primary_share":0.5846,"capability_primary_group_count":17,"capability_non_primary_group_count":26},{"model":"Rombos-LLM-V2.6-Qwen-14B","provider":"Other","country":"XX","openness":"open","release_date":"2024-10-12","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":130,"benchmark_group_count":43,"benchmark_coverage_score":8.98,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.39,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.9462,"source_primary_group_share":0.8372,"capability_breadth":2.25,"capability_primary":"knowledge_general","capability_primary_share":0.5846,"capability_primary_group_count":17,"capability_non_primary_group_count":26},{"model":"Qwen2.5-7B-Instruct-abliterated-v2","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-22","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":130,"benchmark_group_count":43,"benchmark_coverage_score":8.98,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.39,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.9462,"source_primary_group_share":0.8372,"capability_breadth":2.25,"capability_primary":"knowledge_general","capability_primary_share":0.5846,"capability_primary_group_count":17,"capability_non_primary_group_count":26},{"model":"Qwen2.5-7B-Instruct-abliterated","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-19","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":130,"benchmark_group_count":43,"benchmark_coverage_score":8.98,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.39,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.9462,"source_primary_group_share":0.8372,"capability_breadth":2.25,"capability_primary":"knowledge_general","capability_primary_share":0.5846,"capability_primary_group_count":17,"capability_non_primary_group_count":26},{"model":"Qwen2.5-7B-Instruct-Uncensored","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-26","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":130,"benchmark_group_count":43,"benchmark_coverage_score":8.98,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.39,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.9462,"source_primary_group_share":0.8372,"capability_breadth":2.25,"capability_primary":"knowledge_general","capability_primary_share":0.5846,"capability_primary_group_count":17,"capability_non_primary_group_count":26},{"model":"T.E-8.1","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-27","param_count":7.6,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":130,"benchmark_group_count":43,"benchmark_coverage_score":8.98,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.39,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.9462,"source_primary_group_share":0.8372,"capability_breadth":2.25,"capability_primary":"knowledge_general","capability_primary_share":0.5846,"capability_primary_group_count":17,"capability_non_primary_group_count":26},{"model":"Josiefied-Qwen2.5-7B-Instruct-abliterated-v2","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-20","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":130,"benchmark_group_count":43,"benchmark_coverage_score":8.98,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.39,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.9462,"source_primary_group_share":0.8372,"capability_breadth":2.25,"capability_primary":"knowledge_general","capability_primary_share":0.5846,"capability_primary_group_count":17,"capability_non_primary_group_count":26},{"model":"Phi-3-mini-4K-instruct-cpo-simpo","provider":"Microsoft","country":"US","openness":"open","release_date":"2024-06-24","param_count":3.8,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":130,"benchmark_group_count":43,"benchmark_coverage_score":8.98,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.39,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.9462,"source_primary_group_share":0.8372,"capability_breadth":2.25,"capability_primary":"knowledge_general","capability_primary_share":0.5846,"capability_primary_group_count":17,"capability_non_primary_group_count":26},{"model":"Arcee-Spark","provider":"Arcee AI","country":"US","openness":"open","release_date":"2024-06-22","param_count":7.6,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":130,"benchmark_group_count":43,"benchmark_coverage_score":8.98,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.39,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.9462,"source_primary_group_share":0.8372,"capability_breadth":2.25,"capability_primary":"knowledge_general","capability_primary_share":0.5846,"capability_primary_group_count":17,"capability_non_primary_group_count":26},{"model":"calme-2.1-qwen2.5-72B","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-19","param_count":72.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":130,"benchmark_group_count":43,"benchmark_coverage_score":8.98,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.39,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.9462,"source_primary_group_share":0.8372,"capability_breadth":2.25,"capability_primary":"knowledge_general","capability_primary_share":0.5846,"capability_primary_group_count":17,"capability_non_primary_group_count":26},{"model":"calme-2.2-qwen2.5-72B","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-19","param_count":72.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":130,"benchmark_group_count":43,"benchmark_coverage_score":8.98,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.39,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.9462,"source_primary_group_share":0.8372,"capability_breadth":2.25,"capability_primary":"knowledge_general","capability_primary_share":0.5846,"capability_primary_group_count":17,"capability_non_primary_group_count":26},{"model":"SILMA-Kashif-2B-Instruct-v1.0","provider":"SILMA AI","country":"US","openness":"open","release_date":"2025-01-26","param_count":2.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":130,"benchmark_group_count":43,"benchmark_coverage_score":8.98,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.39,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.9462,"source_primary_group_share":0.8372,"capability_breadth":2.25,"capability_primary":"knowledge_general","capability_primary_share":0.5846,"capability_primary_group_count":17,"capability_non_primary_group_count":26},{"model":"LLaMA-65B","provider":"Meta","country":"US","openness":"open","release_date":"2023-02-24","param_count":65.0,"active_param_count":null,"training_compute_flop":5.5e+23,"num_benchmarks":129,"benchmark_group_count":83,"benchmark_coverage_score":45.0,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":4.64,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.4806,"source_primary_group_share":0.3253,"capability_breadth":2.73,"capability_primary":"knowledge_general","capability_primary_share":0.5814,"capability_primary_group_count":38,"capability_non_primary_group_count":45},{"model":"GPT-5.6 Sol","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":129,"benchmark_group_count":108,"benchmark_coverage_score":81.0,"source_family_count":53,"source_group_family_count":52,"source_effective_breadth":11.88,"source_primary":"vals.ai","source_primary_share":0.2205,"source_primary_group_share":0.2407,"capability_breadth":6.37,"capability_primary":"code","capability_primary_share":0.2222,"capability_primary_group_count":26,"capability_non_primary_group_count":81},{"model":"GLM-4.1V-9B (Thinking)","provider":"Zhipu","country":"CN","openness":"open","release_date":"2025-06-28","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":129,"benchmark_group_count":73,"benchmark_coverage_score":9.59,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":1.96,"source_primary":"euroeval.com","source_primary_share":0.7752,"source_primary_group_share":0.6849,"capability_breadth":2.4,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6124,"capability_primary_group_count":49,"capability_non_primary_group_count":24},{"model":"lambda-qwen2.5-32B-dpo-test","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-22","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":129,"benchmark_group_count":42,"benchmark_coverage_score":8.94,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.9535,"source_primary_group_share":0.8571,"capability_breadth":2.24,"capability_primary":"knowledge_general","capability_primary_share":0.5891,"capability_primary_group_count":16,"capability_non_primary_group_count":26},{"model":"Rombos-LLM-V2.5-Qwen-72B","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-30","param_count":72.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":129,"benchmark_group_count":42,"benchmark_coverage_score":8.94,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.9535,"source_primary_group_share":0.8571,"capability_breadth":2.24,"capability_primary":"knowledge_general","capability_primary_share":0.5891,"capability_primary_group_count":16,"capability_non_primary_group_count":26},{"model":"Ultiima-72B","provider":"Other","country":"XX","openness":"open","release_date":"2025-01-09","param_count":72.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":129,"benchmark_group_count":42,"benchmark_coverage_score":8.94,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.9535,"source_primary_group_share":0.8571,"capability_breadth":2.24,"capability_primary":"knowledge_general","capability_primary_share":0.5891,"capability_primary_group_count":16,"capability_non_primary_group_count":26},{"model":"SauerkrautLM-Nemo-12B-Instruct","provider":"VAGOsolutions","country":"DE","openness":"open","release_date":"2024-07-22","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":129,"benchmark_group_count":42,"benchmark_coverage_score":8.94,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/oall/open-arabic-llm-leaderboard","source_primary_share":0.9535,"source_primary_group_share":0.8571,"capability_breadth":2.24,"capability_primary":"knowledge_general","capability_primary_share":0.5891,"capability_primary_group_count":16,"capability_non_primary_group_count":26},{"model":"Grok 4 Fast (Reasoning)","provider":"xAI","country":"US","openness":"api","release_date":"2025-09-19","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":127,"benchmark_group_count":86,"benchmark_coverage_score":68.0,"source_family_count":25,"source_group_family_count":25,"source_effective_breadth":7.04,"source_primary":"artificialanalysis.ai","source_primary_share":0.416,"source_primary_group_share":0.2442,"capability_breadth":6.56,"capability_primary":"knowledge_general","capability_primary_share":0.232,"capability_primary_group_count":18,"capability_non_primary_group_count":68},{"model":"DeepSeek V3.2 Speciale","provider":"DeepSeek","country":"CN","openness":"api","release_date":"2025-12-01","param_count":671.0,"active_param_count":37.0,"training_compute_flop":4.2e+24,"num_benchmarks":127,"benchmark_group_count":84,"benchmark_coverage_score":66.0,"source_family_count":20,"source_group_family_count":20,"source_effective_breadth":8.0,"source_primary":"artificialanalysis.ai","source_primary_share":0.4146,"source_primary_group_share":0.2381,"capability_breadth":7.01,"capability_primary":"math","capability_primary_share":0.1951,"capability_primary_group_count":18,"capability_non_primary_group_count":66},{"model":"DeepSeek V3.2 (Thinking)","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-12-01","param_count":671.0,"active_param_count":37.0,"training_compute_flop":4.2e+24,"num_benchmarks":126,"benchmark_group_count":86,"benchmark_coverage_score":64.0,"source_family_count":22,"source_group_family_count":22,"source_effective_breadth":7.08,"source_primary":"artificialanalysis.ai","source_primary_share":0.4435,"source_primary_group_share":0.2558,"capability_breadth":5.6,"capability_primary":"code","capability_primary_share":0.25,"capability_primary_group_count":22,"capability_non_primary_group_count":64},{"model":"Yi 1.5 34B","provider":"01.AI","country":"CN","openness":"open","release_date":"2024-05-13","param_count":34.0,"active_param_count":null,"training_compute_flop":7.344e+23,"num_benchmarks":126,"benchmark_group_count":70,"benchmark_coverage_score":9.66,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":1.88,"source_primary":"euroeval.com","source_primary_share":0.8,"source_primary_group_share":0.7143,"capability_breadth":2.42,"capability_primary":"knowledge_multilingual","capability_primary_share":0.584,"capability_primary_group_count":43,"capability_non_primary_group_count":27},{"model":"Gemini 2.0 Flash Lite (001)","provider":"Google","country":"US","openness":"api","release_date":"2025-02-25","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 2.0 family deprecated in favor of newer Gemini models.","num_benchmarks":126,"benchmark_group_count":74,"benchmark_coverage_score":9.63,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":2.06,"source_primary":"euroeval.com","source_primary_share":0.8,"source_primary_group_share":0.6757,"capability_breadth":2.41,"capability_primary":"knowledge_multilingual","capability_primary_share":0.616,"capability_primary_group_count":47,"capability_non_primary_group_count":27},{"model":"Phi-4 Multimodal Instruct","provider":"Microsoft","country":"US","openness":"open","release_date":"2025-02-26","param_count":5.6,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":126,"benchmark_group_count":104,"benchmark_coverage_score":75.0,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":4.44,"source_primary":"github.com/audiollms/audiobench","source_primary_share":0.4211,"source_primary_group_share":0.4231,"capability_breadth":5.78,"capability_primary":"vision_language","capability_primary_share":0.2632,"capability_primary_group_count":29,"capability_non_primary_group_count":75},{"model":"Doubao-1.5-Pro","provider":"ByteDance","country":"CN","openness":"api","release_date":"2025-01-15","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":125,"benchmark_group_count":60,"benchmark_coverage_score":9.01,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":1.61,"source_primary":"huggingface.co/spaces/opencompass/open_lmm_reasoning_leaderboard","source_primary_share":0.7686,"source_primary_group_share":0.7833,"capability_breadth":2.25,"capability_primary":"vision_language","capability_primary_share":0.6198,"capability_primary_group_count":41,"capability_non_primary_group_count":19},{"model":"Kimi K3","provider":"Moonshot","country":"CN","openness":"api","release_date":"2026-07-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":125,"benchmark_group_count":93,"benchmark_coverage_score":72.0,"source_family_count":32,"source_group_family_count":32,"source_effective_breadth":8.17,"source_primary":"artificialanalysis.ai","source_primary_share":0.2764,"source_primary_group_share":0.2366,"capability_breadth":5.26,"capability_primary":"knowledge_general","capability_primary_share":0.2787,"capability_primary_group_count":20,"capability_non_primary_group_count":72},{"model":"Qwen 3.6 27B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-04-22","param_count":27.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":125,"benchmark_group_count":95,"benchmark_coverage_score":72.0,"source_family_count":25,"source_group_family_count":24,"source_effective_breadth":7.45,"source_primary":"llm-stats.com","source_primary_share":0.2439,"source_primary_group_share":0.2526,"capability_breadth":6.08,"capability_primary":"vision_language","capability_primary_share":0.2602,"capability_primary_group_count":23,"capability_non_primary_group_count":72},{"model":"GPT-5.1 (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-11-12","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":123,"benchmark_group_count":80,"benchmark_coverage_score":58.0,"source_family_count":26,"source_group_family_count":26,"source_effective_breadth":8.18,"source_primary":"artificialanalysis.ai","source_primary_share":0.4545,"source_primary_group_share":0.2625,"capability_breadth":6.5,"capability_primary":"math","capability_primary_share":0.2149,"capability_primary_group_count":22,"capability_non_primary_group_count":58},{"model":"Claude Opus 4.1","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-08-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":123,"benchmark_group_count":112,"benchmark_coverage_score":92.0,"source_family_count":47,"source_group_family_count":47,"source_effective_breadth":11.06,"source_primary":"benchmarklist.com","source_primary_share":0.2377,"source_primary_group_share":0.2589,"capability_breadth":8.26,"capability_primary":"vision_language","capability_primary_share":0.1803,"capability_primary_group_count":20,"capability_non_primary_group_count":92},{"model":"Phi-1.5","provider":"Microsoft","country":"US","openness":"open","release_date":"2023-09-11","param_count":1.3,"active_param_count":null,"training_compute_flop":1.17e+21,"num_benchmarks":123,"benchmark_group_count":69,"benchmark_coverage_score":39.0,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":5.89,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.5082,"source_primary_group_share":0.3333,"capability_breadth":2.29,"capability_primary":"knowledge_general","capability_primary_share":0.6393,"capability_primary_group_count":30,"capability_non_primary_group_count":39},{"model":"GPT-5.4 (xHigh)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-03-06","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":122,"benchmark_group_count":95,"benchmark_coverage_score":68.0,"source_family_count":28,"source_group_family_count":28,"source_effective_breadth":9.07,"source_primary":"artificialanalysis.ai","source_primary_share":0.3,"source_primary_group_share":0.1684,"capability_breadth":5.83,"capability_primary":"math","capability_primary_share":0.2667,"capability_primary_group_count":27,"capability_non_primary_group_count":68},{"model":"Ministral-8B-Instruct-2410","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-10-16","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":122,"benchmark_group_count":82,"benchmark_coverage_score":49.0,"source_family_count":24,"source_group_family_count":24,"source_effective_breadth":6.99,"source_primary":"euroeval.com","source_primary_share":0.3644,"source_primary_group_share":0.2805,"capability_breadth":3.88,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4322,"capability_primary_group_count":33,"capability_non_primary_group_count":49},{"model":"Claude 3.7 Sonnet (20250219) (Thinking)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-02-19","param_count":null,"active_param_count":null,"training_compute_flop":3.35e+25,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 3 family retired or superseded by newer Claude models.","num_benchmarks":122,"benchmark_group_count":71,"benchmark_coverage_score":9.99,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":1.89,"source_primary":"euroeval.com","source_primary_share":0.8264,"source_primary_group_share":0.7042,"capability_breadth":2.5,"capability_primary":"knowledge_multilingual","capability_primary_share":0.595,"capability_primary_group_count":42,"capability_non_primary_group_count":29},{"model":"Qwen 3 VL 30B A3B (Thinking)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-07-01","param_count":30.0,"active_param_count":3.0,"training_compute_flop":null,"num_benchmarks":122,"benchmark_group_count":77,"benchmark_coverage_score":51.0,"source_family_count":13,"source_group_family_count":12,"source_effective_breadth":4.55,"source_primary":"llm-stats.com","source_primary_share":0.3,"source_primary_group_share":0.2987,"capability_breadth":3.92,"capability_primary":"knowledge_general","capability_primary_share":0.3417,"capability_primary_group_count":26,"capability_non_primary_group_count":51},{"model":"Mistral Medium 3","provider":"Mistral","country":"FR","openness":"api","release_date":"2025-05-07","param_count":123.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":121,"benchmark_group_count":82,"benchmark_coverage_score":61.0,"source_family_count":26,"source_group_family_count":26,"source_effective_breadth":9.31,"source_primary":"artificialanalysis.ai","source_primary_share":0.3077,"source_primary_group_share":0.2195,"capability_breadth":5.65,"capability_primary":"knowledge_general","capability_primary_share":0.2991,"capability_primary_group_count":21,"capability_non_primary_group_count":61},{"model":"Qwen 3.6 Max Preview","provider":"Alibaba","country":"CN","openness":"api","release_date":"2026-04-20","param_count":700.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":121,"benchmark_group_count":78,"benchmark_coverage_score":54.0,"source_family_count":28,"source_group_family_count":28,"source_effective_breadth":10.42,"source_primary":"artificialanalysis.ai","source_primary_share":0.2583,"source_primary_group_share":0.1795,"capability_breadth":4.58,"capability_primary":"code","capability_primary_share":0.3361,"capability_primary_group_count":23,"capability_non_primary_group_count":54},{"model":"LLaMA-13B","provider":"Meta","country":"US","openness":"open","release_date":"2023-02-24","param_count":13.0,"active_param_count":null,"training_compute_flop":7.8e+22,"num_benchmarks":121,"benchmark_group_count":76,"benchmark_coverage_score":42.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":4.61,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.5124,"source_primary_group_share":0.3026,"capability_breadth":2.73,"capability_primary":"knowledge_general","capability_primary_share":0.5785,"capability_primary_group_count":34,"capability_non_primary_group_count":42},{"model":"OLMo-2-1124-7B-Instruct","provider":"Allen AI","country":"US","openness":"open","release_date":"2024-11-24","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":121,"benchmark_group_count":68,"benchmark_coverage_score":8.33,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":1.79,"source_primary":"euroeval.com","source_primary_share":0.8264,"source_primary_group_share":0.7353,"capability_breadth":2.08,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6612,"capability_primary_group_count":47,"capability_non_primary_group_count":21},{"model":"GLM-5-Turbo","provider":"Zhipu","country":"CN","openness":"api","release_date":"2026-02-12","param_count":744.0,"active_param_count":40.0,"training_compute_flop":6.84e+24,"num_benchmarks":121,"benchmark_group_count":95,"benchmark_coverage_score":72.0,"source_family_count":13,"source_group_family_count":12,"source_effective_breadth":2.08,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.5603,"source_primary_group_share":0.6737,"capability_breadth":7.31,"capability_primary":"creativity","capability_primary_share":0.2069,"capability_primary_group_count":23,"capability_non_primary_group_count":72},{"model":"Claude Opus 4.5 (Thinking)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-11-24","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":120,"benchmark_group_count":83,"benchmark_coverage_score":61.0,"source_family_count":15,"source_group_family_count":15,"source_effective_breadth":4.77,"source_primary":"artificialanalysis.ai","source_primary_share":0.4483,"source_primary_group_share":0.3253,"capability_breadth":7.11,"capability_primary":"knowledge_general","capability_primary_share":0.1983,"capability_primary_group_count":22,"capability_non_primary_group_count":61},{"model":"Gemini 1.5 Flash-8B (001)","provider":"Google","country":"US","openness":"api","release_date":"2024-10-03","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 1.x family retired or superseded by newer Gemini models.","num_benchmarks":120,"benchmark_group_count":69,"benchmark_coverage_score":9.14,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":1.8,"source_primary":"euroeval.com","source_primary_share":0.8333,"source_primary_group_share":0.7246,"capability_breadth":2.29,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6333,"capability_primary_group_count":46,"capability_non_primary_group_count":23},{"model":"ERNIE 4.5 300B A47B","provider":"Baidu","country":"CN","openness":"open","release_date":"2025-03-16","param_count":300.0,"active_param_count":47.0,"training_compute_flop":2.82e+24,"num_benchmarks":119,"benchmark_group_count":95,"benchmark_coverage_score":74.0,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":2.03,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.5702,"source_primary_group_share":0.6737,"capability_breadth":7.37,"capability_primary":"creativity","capability_primary_share":0.193,"capability_primary_group_count":21,"capability_non_primary_group_count":74},{"model":"Llama 3.2 3B","provider":"Meta","country":"US","openness":"open","release_date":"2024-09-25","param_count":3.0,"active_param_count":null,"training_compute_flop":1.733e+23,"num_benchmarks":119,"benchmark_group_count":83,"benchmark_coverage_score":8.65,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":3.86,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.4274,"source_primary_group_share":0.4458,"capability_breadth":2.16,"capability_primary":"knowledge_multilingual","capability_primary_share":0.641,"capability_primary_group_count":55,"capability_non_primary_group_count":28},{"model":"Gemma 3n E4B","provider":"Google","country":"US","openness":"open","release_date":"2025-06-26","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":119,"benchmark_group_count":58,"benchmark_coverage_score":8.41,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.24,"source_primary":"euroeval.com","source_primary_share":0.8644,"source_primary_group_share":0.8966,"capability_breadth":2.1,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6525,"capability_primary_group_count":44,"capability_non_primary_group_count":14},{"model":"Llama 3.1 405B Instruct FP8","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-23","param_count":405.0,"active_param_count":null,"training_compute_flop":3.8e+25,"num_benchmarks":117,"benchmark_group_count":62,"benchmark_coverage_score":7.5,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":1.45,"source_primary":"euroeval.com","source_primary_share":0.8632,"source_primary_group_share":0.8226,"capability_breadth":1.87,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7094,"capability_primary_group_count":48,"capability_non_primary_group_count":14},{"model":"Hermes 3 - Llama-3.1 70B","provider":"Nous Research","country":"US","openness":"open","release_date":"2024-08-16","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":116,"benchmark_group_count":109,"benchmark_coverage_score":85.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":2.62,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.5856,"source_primary_group_share":0.5872,"capability_breadth":6.25,"capability_primary":"reasoning_logical","capability_primary_share":0.2252,"capability_primary_group_count":24,"capability_non_primary_group_count":85},{"model":"QwQ 32B-Preview","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-11-27","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":115,"benchmark_group_count":81,"benchmark_coverage_score":41.0,"source_family_count":26,"source_group_family_count":26,"source_effective_breadth":4.34,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.4386,"source_primary_group_share":0.4568,"capability_breadth":2.93,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5351,"capability_primary_group_count":40,"capability_non_primary_group_count":41},{"model":"QVQ-72B-Preview","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-12-24","param_count":72.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":115,"benchmark_group_count":60,"benchmark_coverage_score":8.11,"source_family_count":11,"source_group_family_count":10,"source_effective_breadth":1.61,"source_primary":"huggingface.co/spaces/opencompass/open_lmm_reasoning_leaderboard","source_primary_share":0.8378,"source_primary_group_share":0.7833,"capability_breadth":2.03,"capability_primary":"vision_language","capability_primary_share":0.6306,"capability_primary_group_count":40,"capability_non_primary_group_count":20},{"model":"Gemma 2 2B","provider":"Google","country":"US","openness":"open","release_date":"2024-06-27","param_count":2.0,"active_param_count":null,"training_compute_flop":3.12e+22,"num_benchmarks":115,"benchmark_group_count":69,"benchmark_coverage_score":9.21,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":2.46,"source_primary":"euroeval.com","source_primary_share":0.7391,"source_primary_group_share":0.6087,"capability_breadth":2.3,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6174,"capability_primary_group_count":43,"capability_non_primary_group_count":26},{"model":"Kimi K2.5 (Thinking)","provider":"Moonshot","country":"CN","openness":"open","release_date":"2026-01-27","param_count":1000.0,"active_param_count":32.0,"training_compute_flop":5.8e+24,"num_benchmarks":115,"benchmark_group_count":86,"benchmark_coverage_score":65.0,"source_family_count":29,"source_group_family_count":29,"source_effective_breadth":8.18,"source_primary":"artificialanalysis.ai","source_primary_share":0.3158,"source_primary_group_share":0.2093,"capability_breadth":5.99,"capability_primary":"math","capability_primary_share":0.2368,"capability_primary_group_count":21,"capability_non_primary_group_count":65},{"model":"Qwen 1.5 4B Chat","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-02-05","param_count":4.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":114,"benchmark_group_count":93,"benchmark_coverage_score":51.0,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":4.47,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.4386,"source_primary_group_share":0.3978,"capability_breadth":3.59,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4737,"capability_primary_group_count":42,"capability_non_primary_group_count":51},{"model":"GPT-2","provider":"OpenAI","country":"US","openness":"api","release_date":"2019-02-14","param_count":1.5,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":114,"benchmark_group_count":56,"benchmark_coverage_score":9.87,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.45,"source_primary":"euroeval.com","source_primary_share":0.8304,"source_primary_group_share":0.8214,"capability_breadth":2.47,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5804,"capability_primary_group_count":38,"capability_non_primary_group_count":18},{"model":"GPT-5.2 (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-12-11","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":113,"benchmark_group_count":99,"benchmark_coverage_score":69.0,"source_family_count":32,"source_group_family_count":32,"source_effective_breadth":10.53,"source_primary":"matharena.ai","source_primary_share":0.1944,"source_primary_group_share":0.1717,"capability_breadth":5.51,"capability_primary":"reasoning_logical","capability_primary_share":0.2963,"capability_primary_group_count":30,"capability_non_primary_group_count":69},{"model":"Gemini 2.5 Pro (Preview 05-06)","provider":"Google","country":"US","openness":"api","release_date":"2025-05-06","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 2.5 preview snapshot retired or superseded by newer Gemini models.","num_benchmarks":113,"benchmark_group_count":101,"benchmark_coverage_score":61.0,"source_family_count":32,"source_group_family_count":32,"source_effective_breadth":6.61,"source_primary":"crfm.stanford.edu","source_primary_share":0.3551,"source_primary_group_share":0.3465,"capability_breadth":4.76,"capability_primary":"knowledge_general","capability_primary_share":0.3925,"capability_primary_group_count":40,"capability_non_primary_group_count":61},{"model":"Qwen 1.5 1.8B Chat","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-02-05","param_count":1.8,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":113,"benchmark_group_count":92,"benchmark_coverage_score":50.0,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":4.37,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.4425,"source_primary_group_share":0.4022,"capability_breadth":3.59,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4779,"capability_primary_group_count":42,"capability_non_primary_group_count":50},{"model":"Llama 2 13B Base","provider":"Meta","country":"US","openness":"open","release_date":"2023-07-18","param_count":13.0,"active_param_count":null,"training_compute_flop":1.6e+23,"num_benchmarks":113,"benchmark_group_count":74,"benchmark_coverage_score":10.29,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":3.19,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.4464,"source_primary_group_share":0.5,"capability_breadth":2.57,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5893,"capability_primary_group_count":49,"capability_non_primary_group_count":25},{"model":"GLM-4.5","provider":"Zhipu","country":"CN","openness":"open","release_date":"2025-07-28","param_count":355.0,"active_param_count":32.0,"training_compute_flop":4.42e+24,"num_benchmarks":112,"benchmark_group_count":93,"benchmark_coverage_score":75.0,"source_family_count":38,"source_group_family_count":38,"source_effective_breadth":17.19,"source_primary":"vals.ai","source_primary_share":0.1204,"source_primary_group_share":0.1398,"capability_breadth":7.48,"capability_primary":"knowledge_general","capability_primary_share":0.213,"capability_primary_group_count":18,"capability_non_primary_group_count":75},{"model":"Qwen 1.5 0.5B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-02-04","param_count":0.5,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":112,"benchmark_group_count":77,"benchmark_coverage_score":7.91,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.97,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.4464,"source_primary_group_share":0.4805,"capability_breadth":1.98,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6875,"capability_primary_group_count":56,"capability_non_primary_group_count":21},{"model":"OLMo-2-1124-7B","provider":"Allen AI","country":"US","openness":"open","release_date":"2024-11-24","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":112,"benchmark_group_count":59,"benchmark_coverage_score":7.35,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.36,"source_primary":"euroeval.com","source_primary_share":0.8929,"source_primary_group_share":0.8475,"capability_breadth":1.84,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7143,"capability_primary_group_count":47,"capability_non_primary_group_count":12},{"model":"Grok Code Fast 1","provider":"xAI","country":"US","openness":"api","release_date":"2025-08-28","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":112,"benchmark_group_count":85,"benchmark_coverage_score":50.0,"source_family_count":21,"source_group_family_count":21,"source_effective_breadth":8.78,"source_primary":"artificialanalysis.ai","source_primary_share":0.3238,"source_primary_group_share":0.2,"capability_breadth":4.05,"capability_primary":"code","capability_primary_share":0.4286,"capability_primary_group_count":35,"capability_non_primary_group_count":50},{"model":"Qwen 2.5 Coder 32B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-11-12","param_count":32.0,"active_param_count":null,"training_compute_flop":1.073e+24,"num_benchmarks":111,"benchmark_group_count":98,"benchmark_coverage_score":62.0,"source_family_count":26,"source_group_family_count":26,"source_effective_breadth":5.24,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.4505,"source_primary_group_share":0.3776,"capability_breadth":4.01,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4144,"capability_primary_group_count":36,"capability_non_primary_group_count":62},{"model":"Phi-2","provider":"Microsoft","country":"US","openness":"open","release_date":"2023-12-12","param_count":2.7,"active_param_count":null,"training_compute_flop":2.27e+22,"num_benchmarks":111,"benchmark_group_count":74,"benchmark_coverage_score":55.0,"source_family_count":20,"source_group_family_count":19,"source_effective_breadth":10.25,"source_primary":"euroeval.com","source_primary_share":0.2364,"source_primary_group_share":0.2027,"capability_breadth":5.5,"capability_primary":"knowledge_general","capability_primary_share":0.2636,"capability_primary_group_count":19,"capability_non_primary_group_count":55},{"model":"MiMo-V2-Flash","provider":"Xiaomi","country":"CN","openness":"open","release_date":"2026-01-01","param_count":309.0,"active_param_count":15.0,"training_compute_flop":2.43e+24,"num_benchmarks":111,"benchmark_group_count":90,"benchmark_coverage_score":69.0,"source_family_count":29,"source_group_family_count":29,"source_effective_breadth":12.2,"source_primary":"ppbench.com","source_primary_share":0.1604,"source_primary_group_share":0.1778,"capability_breadth":6.91,"capability_primary":"knowledge_general","capability_primary_share":0.2547,"capability_primary_group_count":21,"capability_non_primary_group_count":69},{"model":"gpt-j-6B","provider":"EleutherAI","country":"US","openness":"open","release_date":"2021-06-09","param_count":6.0,"active_param_count":null,"training_compute_flop":1.5e+22,"num_benchmarks":110,"benchmark_group_count":66,"benchmark_coverage_score":35.0,"source_family_count":8,"source_group_family_count":7,"source_effective_breadth":3.36,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.5636,"source_primary_group_share":0.3939,"capability_breadth":2.3,"capability_primary":"knowledge_general","capability_primary_share":0.6364,"capability_primary_group_count":31,"capability_non_primary_group_count":35},{"model":"LLaMA-7B","provider":"Meta","country":"US","openness":"open","release_date":"2023-02-24","param_count":7.0,"active_param_count":null,"training_compute_flop":4e+22,"num_benchmarks":110,"benchmark_group_count":68,"benchmark_coverage_score":35.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.84,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.5636,"source_primary_group_share":0.3382,"capability_breadth":2.36,"capability_primary":"knowledge_general","capability_primary_share":0.6273,"capability_primary_group_count":33,"capability_non_primary_group_count":35},{"model":"Qwen 1.5 7B Chat","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-02-05","param_count":7.0,"active_param_count":null,"training_compute_flop":1.68e+23,"num_benchmarks":110,"benchmark_group_count":84,"benchmark_coverage_score":46.0,"source_family_count":16,"source_group_family_count":16,"source_effective_breadth":3.92,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.4545,"source_primary_group_share":0.4405,"capability_breadth":3.71,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4636,"capability_primary_group_count":38,"capability_non_primary_group_count":46},{"model":"GPT-5.2 (xHigh)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-12-11","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":109,"benchmark_group_count":85,"benchmark_coverage_score":66.0,"source_family_count":24,"source_group_family_count":24,"source_effective_breadth":8.8,"source_primary":"artificialanalysis.ai","source_primary_share":0.3208,"source_primary_group_share":0.2,"capability_breadth":6.27,"capability_primary":"knowledge_general","capability_primary_share":0.1981,"capability_primary_group_count":19,"capability_non_primary_group_count":66},{"model":"Qwen 3 VL 235B A22B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-09-22","param_count":235.0,"active_param_count":22.0,"training_compute_flop":null,"num_benchmarks":109,"benchmark_group_count":73,"benchmark_coverage_score":45.0,"source_family_count":17,"source_group_family_count":17,"source_effective_breadth":5.74,"source_primary":"llm-stats.com","source_primary_share":0.3333,"source_primary_group_share":0.3014,"capability_breadth":4.75,"capability_primary":"vision_language","capability_primary_share":0.3704,"capability_primary_group_count":28,"capability_non_primary_group_count":45},{"model":"Phi-4-mini (Reasoning)","provider":"Microsoft","country":"US","openness":"open","release_date":"2025-04-30","param_count":3.8,"active_param_count":null,"training_compute_flop":9.956e+22,"num_benchmarks":109,"benchmark_group_count":56,"benchmark_coverage_score":8.51,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.25,"source_primary":"euroeval.com","source_primary_share":0.9174,"source_primary_group_share":0.8929,"capability_breadth":2.13,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6514,"capability_primary_group_count":41,"capability_non_primary_group_count":15},{"model":"zephyr-7B-beta","provider":"HuggingFace","country":"FR","openness":"open","release_date":"2023-10-27","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":108,"benchmark_group_count":90,"benchmark_coverage_score":63.0,"source_family_count":16,"source_group_family_count":16,"source_effective_breadth":5.42,"source_primary":"huggingface.co","source_primary_share":0.3302,"source_primary_group_share":0.3444,"capability_breadth":5.23,"capability_primary":"reasoning_logical","capability_primary_share":0.2642,"capability_primary_group_count":27,"capability_non_primary_group_count":63},{"model":"Claude Sonnet 4 (Thinking)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-05-22","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 4.0 family deprecated in favor of newer Claude 4.x models.","num_benchmarks":108,"benchmark_group_count":76,"benchmark_coverage_score":59.0,"source_family_count":26,"source_group_family_count":26,"source_effective_breadth":9.23,"source_primary":"artificialanalysis.ai","source_primary_share":0.3679,"source_primary_group_share":0.25,"capability_breadth":6.15,"capability_primary":"code","capability_primary_share":0.2642,"capability_primary_group_count":17,"capability_non_primary_group_count":59},{"model":"Jamba-v0.1","provider":"AI21 Labs","country":"IL","openness":"api","release_date":"2024-03-28","param_count":52.0,"active_param_count":12.0,"training_compute_flop":null,"num_benchmarks":108,"benchmark_group_count":58,"benchmark_coverage_score":8.47,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.33,"source_primary":"euroeval.com","source_primary_share":0.9259,"source_primary_group_share":0.8621,"capability_breadth":2.12,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6574,"capability_primary_group_count":41,"capability_non_primary_group_count":17},{"model":"GPT-5.1 (Medium)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-11-12","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":107,"benchmark_group_count":78,"benchmark_coverage_score":43.0,"source_family_count":18,"source_group_family_count":18,"source_effective_breadth":4.28,"source_primary":"medmarks.ai","source_primary_share":0.4175,"source_primary_group_share":0.4231,"capability_breadth":3.35,"capability_primary":"knowledge_general","capability_primary_share":0.4951,"capability_primary_group_count":35,"capability_non_primary_group_count":43},{"model":"Claude 3 Opus (20240229)","provider":"Anthropic","country":"US","openness":"api","release_date":"2024-02-29","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 3 family retired or superseded by newer Claude models.","num_benchmarks":107,"benchmark_group_count":91,"benchmark_coverage_score":76.0,"source_family_count":24,"source_group_family_count":24,"source_effective_breadth":10.19,"source_primary":"crfm.stanford.edu","source_primary_share":0.2075,"source_primary_group_share":0.1868,"capability_breadth":7.47,"capability_primary":"reasoning_logical","capability_primary_share":0.1887,"capability_primary_group_count":15,"capability_non_primary_group_count":76},{"model":"gpt-neox-20B","provider":"EleutherAI","country":"US","openness":"open","release_date":"2022-02-02","param_count":20.0,"active_param_count":null,"training_compute_flop":9.316e+22,"num_benchmarks":107,"benchmark_group_count":64,"benchmark_coverage_score":34.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.2,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.5794,"source_primary_group_share":0.4062,"capability_breadth":2.24,"capability_primary":"knowledge_general","capability_primary_share":0.6449,"capability_primary_group_count":30,"capability_non_primary_group_count":34},{"model":"MiMo-V2.5","provider":"Xiaomi","country":"CN","openness":"open","release_date":"2026-04-23","param_count":310.0,"active_param_count":15.0,"training_compute_flop":null,"num_benchmarks":107,"benchmark_group_count":69,"benchmark_coverage_score":49.0,"source_family_count":24,"source_group_family_count":22,"source_effective_breadth":7.31,"source_primary":"artificialanalysis.ai","source_primary_share":0.3333,"source_primary_group_share":0.2464,"capability_breadth":5.1,"capability_primary":"code","capability_primary_share":0.3048,"capability_primary_group_count":20,"capability_non_primary_group_count":49},{"model":"falcon-7B Instruct","provider":"TII","country":"UAE","openness":"open","release_date":"2023-05-25","param_count":7.0,"active_param_count":null,"training_compute_flop":6.3e+22,"num_benchmarks":106,"benchmark_group_count":50,"benchmark_coverage_score":8.84,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":3.75,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.5905,"source_primary_group_share":0.46,"capability_breadth":2.21,"capability_primary":"knowledge_general","capability_primary_share":0.6571,"capability_primary_group_count":24,"capability_non_primary_group_count":26},{"model":"Qwen 3 30B A3B 2507 (Thinking)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-07-01","param_count":30.0,"active_param_count":3.0,"training_compute_flop":6.48e+23,"num_benchmarks":106,"benchmark_group_count":76,"benchmark_coverage_score":57.0,"source_family_count":20,"source_group_family_count":20,"source_effective_breadth":8.73,"source_primary":"artificialanalysis.ai","source_primary_share":0.3619,"source_primary_group_share":0.25,"capability_breadth":5.86,"capability_primary":"knowledge_general","capability_primary_share":0.2667,"capability_primary_group_count":19,"capability_non_primary_group_count":57},{"model":"Claude Sonnet 5","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-06-30","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":106,"benchmark_group_count":96,"benchmark_coverage_score":72.0,"source_family_count":48,"source_group_family_count":48,"source_effective_breadth":12.22,"source_primary":"vals.ai","source_primary_share":0.2286,"source_primary_group_share":0.2292,"capability_breadth":5.94,"capability_primary":"knowledge_general","capability_primary_share":0.2857,"capability_primary_group_count":24,"capability_non_primary_group_count":72},{"model":"Gemma 3 1B","provider":"Google","country":"US","openness":"open","release_date":"2025-03-12","param_count":1.0,"active_param_count":null,"training_compute_flop":1.2e+22,"num_benchmarks":106,"benchmark_group_count":61,"benchmark_coverage_score":11.24,"source_family_count":9,"source_group_family_count":8,"source_effective_breadth":2.09,"source_primary":"euroeval.com","source_primary_share":0.7358,"source_primary_group_share":0.6721,"capability_breadth":2.81,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5283,"capability_primary_group_count":35,"capability_non_primary_group_count":26},{"model":"miqu-1-70B-sf","provider":"Other","country":"XX","openness":"open","release_date":"2024-02-01","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":106,"benchmark_group_count":56,"benchmark_coverage_score":8.17,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.24,"source_primary":"euroeval.com","source_primary_share":0.9434,"source_primary_group_share":0.8929,"capability_breadth":2.04,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6698,"capability_primary_group_count":41,"capability_non_primary_group_count":15},{"model":"Hermes 4 70B","provider":"Nous Research","country":"US","openness":"open","release_date":"2025-08-26","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":106,"benchmark_group_count":93,"benchmark_coverage_score":66.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":1.9,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.6436,"source_primary_group_share":0.6882,"capability_breadth":4.89,"capability_primary":"knowledge_general","capability_primary_share":0.3267,"capability_primary_group_count":27,"capability_non_primary_group_count":66},{"model":"Claude 3 Haiku (20240307)","provider":"Anthropic","country":"US","openness":"api","release_date":"2024-03-07","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 3 family retired or superseded by newer Claude models.","num_benchmarks":105,"benchmark_group_count":90,"benchmark_coverage_score":74.0,"source_family_count":24,"source_group_family_count":24,"source_effective_breadth":10.36,"source_primary":"crfm.stanford.edu","source_primary_share":0.1845,"source_primary_group_share":0.1889,"capability_breadth":7.47,"capability_primary":"knowledge_multilingual","capability_primary_share":0.1942,"capability_primary_group_count":16,"capability_non_primary_group_count":74},{"model":"Mixtral 8x7B (v0.1)","provider":"Mistral","country":"FR","openness":"open","release_date":"2023-12-11","param_count":46.7,"active_param_count":12.9,"training_compute_flop":7.74e+23,"num_benchmarks":105,"benchmark_group_count":64,"benchmark_coverage_score":39.0,"source_family_count":16,"source_group_family_count":16,"source_effective_breadth":5.57,"source_primary":"euroeval.com","source_primary_share":0.4423,"source_primary_group_share":0.3594,"capability_breadth":3.78,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4135,"capability_primary_group_count":25,"capability_non_primary_group_count":39},{"model":"Qwen 3 VL 4B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-07-01","param_count":4.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":104,"benchmark_group_count":68,"benchmark_coverage_score":39.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":4.37,"source_primary":"artificialanalysis.ai","source_primary_share":0.3398,"source_primary_group_share":0.3088,"capability_breadth":4.65,"capability_primary":"vision_language","capability_primary_share":0.3786,"capability_primary_group_count":29,"capability_non_primary_group_count":39},{"model":"GPT-5.6 Terra","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":104,"benchmark_group_count":90,"benchmark_coverage_score":70.0,"source_family_count":38,"source_group_family_count":38,"source_effective_breadth":8.94,"source_primary":"vals.ai","source_primary_share":0.2621,"source_primary_group_share":0.2778,"capability_breadth":6.54,"capability_primary":"knowledge_general","capability_primary_share":0.2524,"capability_primary_group_count":20,"capability_non_primary_group_count":70},{"model":"Muse Spark","provider":"Meta","country":"US","openness":"private","release_date":"2026-04-08","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":103,"benchmark_group_count":79,"benchmark_coverage_score":54.0,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":6.75,"source_primary":"artificialanalysis.ai","source_primary_share":0.3529,"source_primary_group_share":0.2025,"capability_breadth":4.64,"capability_primary":"knowledge_general","capability_primary_share":0.3267,"capability_primary_group_count":24,"capability_non_primary_group_count":54},{"model":"Llama 2 70B Chat","provider":"Meta","country":"US","openness":"open","release_date":"2023-07-18","param_count":70.0,"active_param_count":null,"training_compute_flop":8.1e+23,"num_benchmarks":103,"benchmark_group_count":58,"benchmark_coverage_score":32.0,"source_family_count":16,"source_group_family_count":16,"source_effective_breadth":4.9,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.6019,"source_primary_group_share":0.3966,"capability_breadth":2.37,"capability_primary":"knowledge_general","capability_primary_share":0.6311,"capability_primary_group_count":26,"capability_non_primary_group_count":32},{"model":"GPT-5.6 Luna","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":103,"benchmark_group_count":88,"benchmark_coverage_score":68.0,"source_family_count":36,"source_group_family_count":36,"source_effective_breadth":8.57,"source_primary":"vals.ai","source_primary_share":0.2673,"source_primary_group_share":0.2841,"capability_breadth":6.49,"capability_primary":"knowledge_general","capability_primary_share":0.2574,"capability_primary_group_count":20,"capability_non_primary_group_count":68},{"model":"O3 Mini (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-01-31","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":102,"benchmark_group_count":76,"benchmark_coverage_score":59.0,"source_family_count":35,"source_group_family_count":35,"source_effective_breadth":11.28,"source_primary":"artificialanalysis.ai","source_primary_share":0.3737,"source_primary_group_share":0.2368,"capability_breadth":6.04,"capability_primary":"code","capability_primary_share":0.2525,"capability_primary_group_count":17,"capability_non_primary_group_count":59},{"model":"Gemma 1.1 7B (IT)","provider":"Google","country":"US","openness":"open","release_date":"2024-04-09","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":102,"benchmark_group_count":77,"benchmark_coverage_score":37.0,"source_family_count":15,"source_group_family_count":15,"source_effective_breadth":3.44,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.4902,"source_primary_group_share":0.4805,"capability_breadth":3.16,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5098,"capability_primary_group_count":40,"capability_non_primary_group_count":37},{"model":"pythia-12B","provider":"EleutherAI","country":"US","openness":"open","release_date":"2023-02-28","param_count":12.0,"active_param_count":null,"training_compute_flop":2.159e+22,"num_benchmarks":102,"benchmark_group_count":60,"benchmark_coverage_score":34.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":3.27,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.6078,"source_primary_group_share":0.3833,"capability_breadth":2.49,"capability_primary":"knowledge_general","capability_primary_share":0.6078,"capability_primary_group_count":26,"capability_non_primary_group_count":34},{"model":"Qwen 3 VL 32B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-07-01","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":102,"benchmark_group_count":70,"benchmark_coverage_score":33.0,"source_family_count":10,"source_group_family_count":9,"source_effective_breadth":4.89,"source_primary":"artificialanalysis.ai","source_primary_share":0.3465,"source_primary_group_share":0.3,"capability_breadth":3.72,"capability_primary":"vision_language","capability_primary_share":0.4554,"capability_primary_group_count":37,"capability_non_primary_group_count":33},{"model":"Claude 3.7 Sonnet (Thinking)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-02-24","param_count":null,"active_param_count":null,"training_compute_flop":3.35e+25,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 3 family retired or superseded by newer Claude models.","num_benchmarks":101,"benchmark_group_count":76,"benchmark_coverage_score":63.0,"source_family_count":27,"source_group_family_count":27,"source_effective_breadth":9.86,"source_primary":"artificialanalysis.ai","source_primary_share":0.3535,"source_primary_group_share":0.2368,"capability_breadth":7.06,"capability_primary":"code","capability_primary_share":0.2121,"capability_primary_group_count":13,"capability_non_primary_group_count":63},{"model":"LLaMA-30B","provider":"Meta","country":"US","openness":"open","release_date":"2023-04-04","param_count":30.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":101,"benchmark_group_count":57,"benchmark_coverage_score":32.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":3.03,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.6139,"source_primary_group_share":0.4035,"capability_breadth":2.53,"capability_primary":"knowledge_general","capability_primary_share":0.604,"capability_primary_group_count":25,"capability_non_primary_group_count":32},{"model":"GPT-2 Medium","provider":"OpenAI","country":"US","openness":"api","release_date":"2019-05-03","param_count":0.355,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":101,"benchmark_group_count":52,"benchmark_coverage_score":8.41,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.26,"source_primary":"euroeval.com","source_primary_share":0.9394,"source_primary_group_share":0.8846,"capability_breadth":2.1,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6566,"capability_primary_group_count":38,"capability_non_primary_group_count":14},{"model":"Devstral Small 2","provider":"Mistral","country":"FR","openness":"open","release_date":"2025-12-09","param_count":24.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":100,"benchmark_group_count":60,"benchmark_coverage_score":44.0,"source_family_count":14,"source_group_family_count":14,"source_effective_breadth":4.7,"source_primary":"artificialanalysis.ai","source_primary_share":0.5789,"source_primary_group_share":0.35,"capability_breadth":5.74,"capability_primary":"code","capability_primary_share":0.2632,"capability_primary_group_count":16,"capability_non_primary_group_count":44},{"model":"DeepSeek V3 Chat","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2024-12-26","param_count":671.0,"active_param_count":37.0,"training_compute_flop":3.3e+24,"num_benchmarks":100,"benchmark_group_count":90,"benchmark_coverage_score":66.0,"source_family_count":18,"source_group_family_count":18,"source_effective_breadth":1.95,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.6989,"source_primary_group_share":0.7111,"capability_breadth":6.6,"capability_primary":"creativity","capability_primary_share":0.2688,"capability_primary_group_count":24,"capability_non_primary_group_count":66},{"model":"Mistral 7B Instruct (v0.1)","provider":"Mistral","country":"FR","openness":"open","release_date":"2023-09-27","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":100,"benchmark_group_count":66,"benchmark_coverage_score":41.0,"source_family_count":20,"source_group_family_count":20,"source_effective_breadth":7.24,"source_primary":"euroeval.com","source_primary_share":0.37,"source_primary_group_share":0.303,"capability_breadth":3.99,"capability_primary":"knowledge_multilingual","capability_primary_share":0.42,"capability_primary_group_count":25,"capability_non_primary_group_count":41},{"model":"vicuna-7B-v1.5","provider":"LMSYS","country":"US","openness":"open","release_date":"2023-07-29","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":100,"benchmark_group_count":53,"benchmark_coverage_score":33.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":3.25,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.62,"source_primary_group_share":0.434,"capability_breadth":2.29,"capability_primary":"knowledge_general","capability_primary_share":0.64,"capability_primary_group_count":20,"capability_non_primary_group_count":33},{"model":"GLM-4.5V","provider":"Zhipu","country":"CN","openness":"open","release_date":"2025-07-28","param_count":106.0,"active_param_count":12.0,"training_compute_flop":1.8e+24,"num_benchmarks":100,"benchmark_group_count":93,"benchmark_coverage_score":72.0,"source_family_count":14,"source_group_family_count":14,"source_effective_breadth":2.1,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.6598,"source_primary_group_share":0.6774,"capability_breadth":6.52,"capability_primary":"creativity","capability_primary_share":0.2268,"capability_primary_group_count":21,"capability_non_primary_group_count":72},{"model":"gpt2-large","provider":"OpenAI","country":"US","openness":"api","release_date":"2019-08-20","param_count":0.774,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":100,"benchmark_group_count":52,"benchmark_coverage_score":8.41,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.26,"source_primary":"euroeval.com","source_primary_share":0.9394,"source_primary_group_share":0.8846,"capability_breadth":2.1,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6566,"capability_primary_group_count":38,"capability_non_primary_group_count":14},{"model":"mpt-30B","provider":"Databricks","country":"US","openness":"open","release_date":"2023-06-22","param_count":30.0,"active_param_count":null,"training_compute_flop":1.89e+23,"num_benchmarks":98,"benchmark_group_count":57,"benchmark_coverage_score":8.12,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.96,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.6327,"source_primary_group_share":0.4035,"capability_breadth":2.03,"capability_primary":"knowledge_general","capability_primary_share":0.6837,"capability_primary_group_count":31,"capability_non_primary_group_count":26},{"model":"Llama 4 Maverick Instruct FP8","provider":"Meta","country":"US","openness":"open","release_date":"2025-04-05","param_count":400.0,"active_param_count":17.0,"training_compute_flop":2.244e+24,"num_benchmarks":98,"benchmark_group_count":84,"benchmark_coverage_score":60.0,"source_family_count":15,"source_group_family_count":15,"source_effective_breadth":3.97,"source_primary":"crfm.stanford.edu","source_primary_share":0.4167,"source_primary_group_share":0.4643,"capability_breadth":7.11,"capability_primary":"knowledge_multilingual","capability_primary_share":0.25,"capability_primary_group_count":24,"capability_non_primary_group_count":60},{"model":"Qwen 3 VL 8B (Thinking)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-07-01","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":97,"benchmark_group_count":64,"benchmark_coverage_score":32.0,"source_family_count":14,"source_group_family_count":13,"source_effective_breadth":4.18,"source_primary":"llm-stats.com","source_primary_share":0.375,"source_primary_group_share":0.3594,"capability_breadth":3.97,"capability_primary":"vision_language","capability_primary_share":0.4375,"capability_primary_group_count":32,"capability_non_primary_group_count":32},{"model":"O1 Mini","provider":"OpenAI","country":"US","openness":"api","release_date":"2024-09-12","param_count":100.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI o1 family deprecated or superseded by newer o-series models.","num_benchmarks":97,"benchmark_group_count":84,"benchmark_coverage_score":62.0,"source_family_count":44,"source_group_family_count":44,"source_effective_breadth":15.96,"source_primary":"benchmarklist.com","source_primary_share":0.1667,"source_primary_group_share":0.1905,"capability_breadth":6.74,"capability_primary":"code","capability_primary_share":0.2292,"capability_primary_group_count":22,"capability_non_primary_group_count":62},{"model":"Mistral-7B-v0.3","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-05-22","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":97,"benchmark_group_count":61,"benchmark_coverage_score":31.0,"source_family_count":17,"source_group_family_count":17,"source_effective_breadth":6.07,"source_primary":"euroeval.com","source_primary_share":0.3814,"source_primary_group_share":0.3279,"capability_breadth":3.04,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5155,"capability_primary_group_count":30,"capability_non_primary_group_count":31},{"model":"Kimi K2.7 Code","provider":"Moonshot","country":"CN","openness":"api","release_date":"2026-06-12","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":97,"benchmark_group_count":73,"benchmark_coverage_score":44.0,"source_family_count":31,"source_group_family_count":31,"source_effective_breadth":11.51,"source_primary":"artificialanalysis.ai","source_primary_share":0.3617,"source_primary_group_share":0.2055,"capability_breadth":4.15,"capability_primary":"code","capability_primary_share":0.4086,"capability_primary_group_count":28,"capability_non_primary_group_count":44},{"model":"Nova Pro","provider":"Amazon","country":"US","openness":"api","release_date":"2024-12-03","param_count":90.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":96,"benchmark_group_count":70,"benchmark_coverage_score":54.0,"source_family_count":16,"source_group_family_count":16,"source_effective_breadth":6.71,"source_primary":"artificialanalysis.ai","source_primary_share":0.375,"source_primary_group_share":0.2571,"capability_breadth":5.6,"capability_primary":"knowledge_general","capability_primary_share":0.3125,"capability_primary_group_count":16,"capability_non_primary_group_count":54},{"model":"Phi-3-medium-4k-instruct","provider":"Microsoft","country":"US","openness":"open","release_date":"2024-05-21","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":96,"benchmark_group_count":77,"benchmark_coverage_score":50.0,"source_family_count":17,"source_group_family_count":17,"source_effective_breadth":5.85,"source_primary":"huggingface.co","source_primary_share":0.2917,"source_primary_group_share":0.3117,"capability_breadth":4.3,"capability_primary":"knowledge_multilingual","capability_primary_share":0.375,"capability_primary_group_count":27,"capability_non_primary_group_count":50},{"model":"GPT-5.5 (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-04-23","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":96,"benchmark_group_count":66,"benchmark_coverage_score":46.0,"source_family_count":29,"source_group_family_count":29,"source_effective_breadth":10.42,"source_primary":"artificialanalysis.ai","source_primary_share":0.3871,"source_primary_group_share":0.2424,"capability_breadth":5.17,"capability_primary":"code","capability_primary_share":0.3261,"capability_primary_group_count":19,"capability_non_primary_group_count":46},{"model":"Mistral Medium 3.5","provider":"Mistral","country":"FR","openness":"api","release_date":"2025-12-02","param_count":123.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":96,"benchmark_group_count":62,"benchmark_coverage_score":46.0,"source_family_count":18,"source_group_family_count":18,"source_effective_breadth":6.56,"source_primary":"artificialanalysis.ai","source_primary_share":0.3789,"source_primary_group_share":0.2581,"capability_breadth":5.43,"capability_primary":"knowledge_general","capability_primary_share":0.3085,"capability_primary_group_count":15,"capability_non_primary_group_count":46},{"model":"Mistral Large","provider":"Mistral","country":"FR","openness":"api","release_date":"2024-02-26","param_count":123.0,"active_param_count":null,"training_compute_flop":1.12e+25,"num_benchmarks":95,"benchmark_group_count":86,"benchmark_coverage_score":66.0,"source_family_count":28,"source_group_family_count":28,"source_effective_breadth":10.72,"source_primary":"livebench.ai","source_primary_share":0.1889,"source_primary_group_share":0.1977,"capability_breadth":7.08,"capability_primary":"knowledge_general","capability_primary_share":0.2444,"capability_primary_group_count":20,"capability_non_primary_group_count":66},{"model":"Mixtral 8x22B Instruct","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-04-17","param_count":141.0,"active_param_count":39.0,"training_compute_flop":2.34e+24,"num_benchmarks":95,"benchmark_group_count":89,"benchmark_coverage_score":67.0,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":1.89,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.7222,"source_primary_group_share":0.7191,"capability_breadth":6.57,"capability_primary":"creativity","capability_primary_share":0.2556,"capability_primary_group_count":22,"capability_non_primary_group_count":67},{"model":"INTELLECT-3","provider":"PrimeIntellect","country":"US","openness":"open","release_date":"2025-12-01","param_count":106.0,"active_param_count":12.0,"training_compute_flop":null,"num_benchmarks":95,"benchmark_group_count":55,"benchmark_coverage_score":30.0,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":3.46,"source_primary":"artificialanalysis.ai","source_primary_share":0.5426,"source_primary_group_share":0.3818,"capability_breadth":3.64,"capability_primary":"knowledge_general","capability_primary_share":0.4574,"capability_primary_group_count":25,"capability_non_primary_group_count":30},{"model":"Claude 3 Sonnet (20240229)","provider":"Anthropic","country":"US","openness":"api","release_date":"2024-03-04","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 3 family retired or superseded by newer Claude models.","num_benchmarks":94,"benchmark_group_count":79,"benchmark_coverage_score":66.0,"source_family_count":19,"source_group_family_count":19,"source_effective_breadth":8.58,"source_primary":"crfm.stanford.edu","source_primary_share":0.2366,"source_primary_group_share":0.2152,"capability_breadth":8.14,"capability_primary":"reasoning_logical","capability_primary_share":0.1828,"capability_primary_group_count":13,"capability_non_primary_group_count":66},{"model":"gpt-sw3-126m-instruct","provider":"AI Sweden","country":"SE","openness":"open","release_date":"2023-04-28","param_count":0.126,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":94,"benchmark_group_count":47,"benchmark_coverage_score":7.38,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.23,"source_primary":"euroeval.com","source_primary_share":0.9255,"source_primary_group_share":0.8936,"capability_breadth":1.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7128,"capability_primary_group_count":38,"capability_non_primary_group_count":9},{"model":"gpt-sw3-356m-instruct","provider":"AI Sweden","country":"SE","openness":"open","release_date":"2023-04-28","param_count":0.356,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":94,"benchmark_group_count":47,"benchmark_coverage_score":7.38,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.23,"source_primary":"euroeval.com","source_primary_share":0.9255,"source_primary_group_share":0.8936,"capability_breadth":1.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7128,"capability_primary_group_count":38,"capability_non_primary_group_count":9},{"model":"Qwen 2 7B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-06-07","param_count":7.0,"active_param_count":null,"training_compute_flop":2.94e+23,"num_benchmarks":93,"benchmark_group_count":67,"benchmark_coverage_score":7.68,"source_family_count":15,"source_group_family_count":15,"source_effective_breadth":3.02,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.5435,"source_primary_group_share":0.5522,"capability_breadth":1.92,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6957,"capability_primary_group_count":47,"capability_non_primary_group_count":20},{"model":"alpaca-7B","provider":"Other","country":"XX","openness":"open","release_date":"2023-03-18","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":93,"benchmark_group_count":52,"benchmark_coverage_score":8.81,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.63,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.4423,"capability_breadth":2.2,"capability_primary":"knowledge_general","capability_primary_share":0.6559,"capability_primary_group_count":25,"capability_non_primary_group_count":27},{"model":"Qwen 3 8B (Thinking)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":8.0,"active_param_count":null,"training_compute_flop":1.771e+24,"num_benchmarks":93,"benchmark_group_count":60,"benchmark_coverage_score":8.74,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.42,"source_primary":"medmarks.ai","source_primary_share":0.4624,"source_primary_group_share":0.55,"capability_breadth":2.18,"capability_primary":"knowledge_general","capability_primary_share":0.6452,"capability_primary_group_count":39,"capability_non_primary_group_count":21},{"model":"nemotron-3-super-120B-a12B","provider":"NVIDIA","country":"US","openness":"open","release_date":"2026-03-11","param_count":120.0,"active_param_count":12.0,"training_compute_flop":null,"num_benchmarks":93,"benchmark_group_count":86,"benchmark_coverage_score":65.0,"source_family_count":18,"source_group_family_count":17,"source_effective_breadth":1.78,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.7303,"source_primary_group_share":0.7442,"capability_breadth":6.68,"capability_primary":"creativity","capability_primary_share":0.2472,"capability_primary_group_count":21,"capability_non_primary_group_count":65},{"model":"Llama 2 7B","provider":"Meta","country":"US","openness":"open","release_date":"2023-07-18","param_count":7.0,"active_param_count":null,"training_compute_flop":8.4e+22,"num_benchmarks":92,"benchmark_group_count":82,"benchmark_coverage_score":56.0,"source_family_count":20,"source_group_family_count":20,"source_effective_breadth":6.28,"source_primary":"crfm.stanford.edu","source_primary_share":0.2935,"source_primary_group_share":0.3293,"capability_breadth":6.58,"capability_primary":"knowledge_general","capability_primary_share":0.2935,"capability_primary_group_count":26,"capability_non_primary_group_count":56},{"model":"starcoder","provider":"BigCode","country":"FR","openness":"open","release_date":"2023-04-24","param_count":15.5,"active_param_count":null,"training_compute_flop":8.46e+22,"num_benchmarks":92,"benchmark_group_count":43,"benchmark_coverage_score":10.68,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.88,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.6889,"source_primary_group_share":0.5349,"capability_breadth":2.67,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":14,"capability_non_primary_group_count":29},{"model":"Dolly V2 12B","provider":"Databricks","country":"US","openness":"open","release_date":"2023-04-12","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":92,"benchmark_group_count":49,"benchmark_coverage_score":9.52,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.48,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.6813,"source_primary_group_share":0.4694,"capability_breadth":2.38,"capability_primary":"knowledge_general","capability_primary_share":0.6264,"capability_primary_group_count":21,"capability_non_primary_group_count":28},{"model":"Qwen 3 30B A3B (Thinking)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":30.0,"active_param_count":3.0,"training_compute_flop":6.48e+23,"num_benchmarks":91,"benchmark_group_count":59,"benchmark_coverage_score":8.82,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":2.44,"source_primary":"medmarks.ai","source_primary_share":0.4778,"source_primary_group_share":0.5593,"capability_breadth":2.2,"capability_primary":"knowledge_general","capability_primary_share":0.6444,"capability_primary_group_count":37,"capability_non_primary_group_count":22},{"model":"GPT-5.3 Codex","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-02-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":91,"benchmark_group_count":80,"benchmark_coverage_score":47.0,"source_family_count":39,"source_group_family_count":39,"source_effective_breadth":11.81,"source_primary":"benchmarklist.com","source_primary_share":0.2273,"source_primary_group_share":0.25,"capability_breadth":3.63,"capability_primary":"code","capability_primary_share":0.3977,"capability_primary_group_count":33,"capability_non_primary_group_count":47},{"model":"Phi-3 Mini 4K Instruct","provider":"Microsoft","country":"US","openness":"open","release_date":"2024-04-23","param_count":3.8,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":90,"benchmark_group_count":76,"benchmark_coverage_score":51.0,"source_family_count":19,"source_group_family_count":19,"source_effective_breadth":8.7,"source_primary":"livebench.ai","source_primary_share":0.191,"source_primary_group_share":0.2237,"capability_breadth":4.77,"capability_primary":"reasoning_logical","capability_primary_share":0.3034,"capability_primary_group_count":25,"capability_non_primary_group_count":51},{"model":"Grok 4.20 0309 (Reasoning)","provider":"xAI","country":"US","openness":"api","release_date":"2026-03-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":90,"benchmark_group_count":56,"benchmark_coverage_score":40.0,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":3.56,"source_primary":"artificialanalysis.ai","source_primary_share":0.5618,"source_primary_group_share":0.3929,"capability_breadth":5.43,"capability_primary":"knowledge_general","capability_primary_share":0.3034,"capability_primary_group_count":16,"capability_non_primary_group_count":40},{"model":"GPT-OSS-120B (Low)","provider":"OpenAI","country":"US","openness":"open","release_date":"2025-08-05","param_count":120.0,"active_param_count":5.13,"training_compute_flop":4.94e+24,"num_benchmarks":90,"benchmark_group_count":50,"benchmark_coverage_score":14.91,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.77,"source_primary":"artificialanalysis.ai","source_primary_share":0.6067,"source_primary_group_share":0.42,"capability_breadth":3.73,"capability_primary":"knowledge_general","capability_primary_share":0.4494,"capability_primary_group_count":22,"capability_non_primary_group_count":28},{"model":"ChatGLM2 6B","provider":"Zhipu","country":"CN","openness":"open","release_date":"2023-06-25","param_count":6.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":89,"benchmark_group_count":49,"benchmark_coverage_score":30.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":3.47,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.7045,"source_primary_group_share":0.4694,"capability_breadth":2.4,"capability_primary":"knowledge_general","capability_primary_share":0.625,"capability_primary_group_count":19,"capability_non_primary_group_count":30},{"model":"c4ai-command-r-08-2024","provider":"Cohere","country":"CA","openness":"open","release_date":"2024-08-01","param_count":35.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":89,"benchmark_group_count":61,"benchmark_coverage_score":40.0,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":3.96,"source_primary":"euroeval.com","source_primary_share":0.5227,"source_primary_group_share":0.377,"capability_breadth":4.55,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3864,"capability_primary_group_count":21,"capability_non_primary_group_count":40},{"model":"CodeLlama-34B-Instruct-hf","provider":"Meta","country":"US","openness":"open","release_date":"2023-08-24","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":89,"benchmark_group_count":44,"benchmark_coverage_score":30.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.84,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.7209,"source_primary_group_share":0.5227,"capability_breadth":2.66,"capability_primary":"knowledge_general","capability_primary_share":0.5814,"capability_primary_group_count":14,"capability_non_primary_group_count":30},{"model":"Claude Sonnet 4 (20250514)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-05-14","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 4.0 family deprecated in favor of newer Claude 4.x models.","num_benchmarks":89,"benchmark_group_count":82,"benchmark_coverage_score":62.0,"source_family_count":27,"source_group_family_count":27,"source_effective_breadth":12.09,"source_primary":"vals.ai","source_primary_share":0.1839,"source_primary_group_share":0.1951,"capability_breadth":7.72,"capability_primary":"knowledge_general","capability_primary_share":0.2299,"capability_primary_group_count":20,"capability_non_primary_group_count":62},{"model":"Grok 4.5","provider":"xAI","country":"US","openness":"api","release_date":"2026-07-08","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":89,"benchmark_group_count":77,"benchmark_coverage_score":57.0,"source_family_count":41,"source_group_family_count":41,"source_effective_breadth":9.4,"source_primary":"vals.ai","source_primary_share":0.2874,"source_primary_group_share":0.2987,"capability_breadth":5.61,"capability_primary":"knowledge_general","capability_primary_share":0.2759,"capability_primary_group_count":20,"capability_non_primary_group_count":57},{"model":"Qwen 2.5 Coder 7B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-19","param_count":7.0,"active_param_count":null,"training_compute_flop":2.511e+23,"num_benchmarks":88,"benchmark_group_count":74,"benchmark_coverage_score":38.0,"source_family_count":17,"source_group_family_count":17,"source_effective_breadth":3.62,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.5747,"source_primary_group_share":0.5,"capability_breadth":3.02,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5287,"capability_primary_group_count":36,"capability_non_primary_group_count":38},{"model":"granite-3.0-8B-base","provider":"IBM","country":"US","openness":"open","release_date":"2024-10-21","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":88,"benchmark_group_count":51,"benchmark_coverage_score":7.03,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.32,"source_primary":"euroeval.com","source_primary_share":0.9205,"source_primary_group_share":0.8627,"capability_breadth":1.76,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7386,"capability_primary_group_count":39,"capability_non_primary_group_count":12},{"model":"vicuna-13B-v1.3","provider":"LMSYS","country":"US","openness":"open","release_date":"2023-06-18","param_count":13.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":87,"benchmark_group_count":44,"benchmark_coverage_score":9.63,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":3.0,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.7126,"source_primary_group_share":0.5227,"capability_breadth":2.41,"capability_primary":"knowledge_general","capability_primary_share":0.6207,"capability_primary_group_count":18,"capability_non_primary_group_count":26},{"model":"Qwen 3 VL 30B A3B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-07-01","param_count":30.0,"active_param_count":3.0,"training_compute_flop":null,"num_benchmarks":87,"benchmark_group_count":55,"benchmark_coverage_score":31.0,"source_family_count":9,"source_group_family_count":8,"source_effective_breadth":3.43,"source_primary":"llm-stats.com","source_primary_share":0.4138,"source_primary_group_share":0.4182,"capability_breadth":4.37,"capability_primary":"vision_language","capability_primary_share":0.3908,"capability_primary_group_count":24,"capability_non_primary_group_count":31},{"model":"falcon-7B","provider":"TII","country":"UAE","openness":"open","release_date":"2023-05-25","param_count":7.0,"active_param_count":null,"training_compute_flop":6.3e+22,"num_benchmarks":86,"benchmark_group_count":62,"benchmark_coverage_score":35.0,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":4.13,"source_primary":"crfm.stanford.edu","source_primary_share":0.3176,"source_primary_group_share":0.4355,"capability_breadth":4.29,"capability_primary":"knowledge_general","capability_primary_share":0.4353,"capability_primary_group_count":27,"capability_non_primary_group_count":35},{"model":"Nova Lite (v1)","provider":"Amazon","country":"US","openness":"api","release_date":"2024-12-03","param_count":20.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":86,"benchmark_group_count":77,"benchmark_coverage_score":55.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.38,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.8148,"source_primary_group_share":0.8442,"capability_breadth":5.82,"capability_primary":"creativity","capability_primary_share":0.284,"capability_primary_group_count":22,"capability_non_primary_group_count":55},{"model":"Qwen 3 VL 4B (Thinking)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-07-01","param_count":4.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":85,"benchmark_group_count":52,"benchmark_coverage_score":16.6,"source_family_count":8,"source_group_family_count":7,"source_effective_breadth":3.24,"source_primary":"artificialanalysis.ai","source_primary_share":0.4167,"source_primary_group_share":0.4231,"capability_breadth":4.15,"capability_primary":"vision_language","capability_primary_share":0.4048,"capability_primary_group_count":24,"capability_non_primary_group_count":28},{"model":"Magistral Medium","provider":"Mistral","country":"FR","openness":"api","release_date":"2025-06-10","param_count":123.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":85,"benchmark_group_count":57,"benchmark_coverage_score":40.0,"source_family_count":16,"source_group_family_count":16,"source_effective_breadth":5.2,"source_primary":"artificialanalysis.ai","source_primary_share":0.4118,"source_primary_group_share":0.3158,"capability_breadth":4.23,"capability_primary":"knowledge_general","capability_primary_share":0.3647,"capability_primary_group_count":17,"capability_non_primary_group_count":40},{"model":"vicuna-7B-v1.3","provider":"LMSYS","country":"US","openness":"open","release_date":"2023-06-18","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":85,"benchmark_group_count":43,"benchmark_coverage_score":8.53,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.88,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.7381,"source_primary_group_share":0.5349,"capability_breadth":2.13,"capability_primary":"knowledge_general","capability_primary_share":0.6667,"capability_primary_group_count":20,"capability_non_primary_group_count":23},{"model":"nemotron-3-nano-30B-a3B","provider":"NVIDIA","country":"US","openness":"open","release_date":"2025-12-15","param_count":30.0,"active_param_count":3.0,"training_compute_flop":4.8e+23,"num_benchmarks":85,"benchmark_group_count":77,"benchmark_coverage_score":55.0,"source_family_count":11,"source_group_family_count":10,"source_effective_breadth":1.44,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.8125,"source_primary_group_share":0.8312,"capability_breadth":5.69,"capability_primary":"creativity","capability_primary_share":0.2875,"capability_primary_group_count":22,"capability_non_primary_group_count":55},{"model":"trinity-large-preview","provider":"Arcee AI","country":"US","openness":"open","release_date":"2026-01-27","param_count":400.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":85,"benchmark_group_count":78,"benchmark_coverage_score":55.0,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":1.48,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.8228,"source_primary_group_share":0.8205,"capability_breadth":5.58,"capability_primary":"creativity","capability_primary_share":0.3038,"capability_primary_group_count":23,"capability_non_primary_group_count":55},{"model":"GPT-3.5 Turbo (0125)","provider":"OpenAI","country":"US","openness":"api","release_date":"2024-01-25","param_count":175.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI GPT-3/GPT-3.5 family retired or superseded by newer API models.","num_benchmarks":84,"benchmark_group_count":69,"benchmark_coverage_score":54.0,"source_family_count":26,"source_group_family_count":26,"source_effective_breadth":9.39,"source_primary":"crfm.stanford.edu","source_primary_share":0.2024,"source_primary_group_share":0.2464,"capability_breadth":6.38,"capability_primary":"knowledge_multilingual","capability_primary_share":0.2262,"capability_primary_group_count":15,"capability_non_primary_group_count":54},{"model":"GPT-5 Mini (Medium)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-08-07","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":84,"benchmark_group_count":53,"benchmark_coverage_score":43.0,"source_family_count":19,"source_group_family_count":19,"source_effective_breadth":6.52,"source_primary":"artificialanalysis.ai","source_primary_share":0.4167,"source_primary_group_share":0.3208,"capability_breadth":6.05,"capability_primary":"knowledge_general","capability_primary_share":0.2262,"capability_primary_group_count":10,"capability_non_primary_group_count":43},{"model":"opt-1.3B","provider":"Meta","country":"US","openness":"open","release_date":"2022-05-03","param_count":1.3,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":84,"benchmark_group_count":42,"benchmark_coverage_score":8.7,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.71,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.7381,"source_primary_group_share":0.5476,"capability_breadth":2.18,"capability_primary":"knowledge_general","capability_primary_share":0.6548,"capability_primary_group_count":19,"capability_non_primary_group_count":23},{"model":"Nemotron 3 Ultra","provider":"NVIDIA","country":"US","openness":"open","release_date":"2026-06-04","param_count":550.0,"active_param_count":55.0,"training_compute_flop":6.6e+24,"num_benchmarks":84,"benchmark_group_count":65,"benchmark_coverage_score":47.0,"source_family_count":21,"source_group_family_count":21,"source_effective_breadth":9.45,"source_primary":"huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","source_primary_share":0.2317,"source_primary_group_share":0.2,"capability_breadth":5.36,"capability_primary":"code","capability_primary_share":0.3293,"capability_primary_group_count":18,"capability_non_primary_group_count":47},{"model":"Hermes 4 405B","provider":"Nous Research","country":"US","openness":"open","release_date":"2025-08-26","param_count":405.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":84,"benchmark_group_count":78,"benchmark_coverage_score":53.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":1.47,"source_primary":"huggingface.co/spaces/agcbench-2026/agc-bench-leaderboard","source_primary_share":0.8228,"source_primary_group_share":0.8205,"capability_breadth":5.15,"capability_primary":"creativity","capability_primary_share":0.3291,"capability_primary_group_count":25,"capability_non_primary_group_count":53},{"model":"Qwen 3 VL 32B (Thinking)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-07-01","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":83,"benchmark_group_count":49,"benchmark_coverage_score":17.2,"source_family_count":7,"source_group_family_count":6,"source_effective_breadth":3.31,"source_primary":"artificialanalysis.ai","source_primary_share":0.4268,"source_primary_group_share":0.3878,"capability_breadth":4.3,"capability_primary":"vision_language","capability_primary_share":0.3902,"capability_primary_group_count":21,"capability_non_primary_group_count":28},{"model":"O1 (2024-12-17)","provider":"OpenAI","country":"US","openness":"api","release_date":"2024-12-17","param_count":200.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI o1 family deprecated or superseded by newer o-series models.","num_benchmarks":83,"benchmark_group_count":74,"benchmark_coverage_score":58.0,"source_family_count":20,"source_group_family_count":20,"source_effective_breadth":8.83,"source_primary":"huggingface.co/spaces/galileo-ai/agent-leaderboard","source_primary_share":0.175,"source_primary_group_share":0.1757,"capability_breadth":7.6,"capability_primary":"agentic","capability_primary_share":0.225,"capability_primary_group_count":16,"capability_non_primary_group_count":58},{"model":"Qwen 2 VL 7B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-08-29","param_count":7.0,"active_param_count":null,"training_compute_flop":6.72e+22,"num_benchmarks":83,"benchmark_group_count":63,"benchmark_coverage_score":10.92,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":2.58,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.6024,"source_primary_group_share":0.5873,"capability_breadth":2.73,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5422,"capability_primary_group_count":35,"capability_non_primary_group_count":28},{"model":"MiMo-V2-Pro","provider":"Xiaomi","country":"CN","openness":"open","release_date":"2026-03-18","param_count":1000.0,"active_param_count":42.0,"training_compute_flop":null,"num_benchmarks":83,"benchmark_group_count":57,"benchmark_coverage_score":43.0,"source_family_count":25,"source_group_family_count":24,"source_effective_breadth":7.37,"source_primary":"artificialanalysis.ai","source_primary_share":0.3735,"source_primary_group_share":0.2456,"capability_breadth":5.27,"capability_primary":"code","capability_primary_share":0.2892,"capability_primary_group_count":14,"capability_non_primary_group_count":43},{"model":"Ling-flash-2.0","provider":"InclusionAI","country":"CN","openness":"open","release_date":"2025-09-17","param_count":100.0,"active_param_count":6.1,"training_compute_flop":null,"num_benchmarks":83,"benchmark_group_count":58,"benchmark_coverage_score":34.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":4.05,"source_primary":"artificialanalysis.ai","source_primary_share":0.4096,"source_primary_group_share":0.3621,"capability_breadth":3.53,"capability_primary":"knowledge_general","capability_primary_share":0.4578,"capability_primary_group_count":24,"capability_non_primary_group_count":34},{"model":"Gemini 3.5 Flash (High)","provider":"Google","country":"US","openness":"api","release_date":"2026-05-19","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":82,"benchmark_group_count":57,"benchmark_coverage_score":44.0,"source_family_count":15,"source_group_family_count":15,"source_effective_breadth":6.33,"source_primary":"artificialanalysis.ai","source_primary_share":0.4625,"source_primary_group_share":0.2982,"capability_breadth":5.73,"capability_primary":"code","capability_primary_share":0.2658,"capability_primary_group_count":12,"capability_non_primary_group_count":44},{"model":"falcon-40B Instruct","provider":"TII","country":"UAE","openness":"open","release_date":"2023-05-25","param_count":40.0,"active_param_count":null,"training_compute_flop":2.4e+23,"num_benchmarks":82,"benchmark_group_count":40,"benchmark_coverage_score":8.43,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.67,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.7561,"source_primary_group_share":0.575,"capability_breadth":2.11,"capability_primary":"knowledge_general","capability_primary_share":0.6707,"capability_primary_group_count":18,"capability_non_primary_group_count":22},{"model":"vicuna-7B-v1.5-16k","provider":"LMSYS","country":"US","openness":"open","release_date":"2023-07-31","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":82,"benchmark_group_count":43,"benchmark_coverage_score":9.88,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.24,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.7561,"source_primary_group_share":0.5349,"capability_breadth":2.47,"capability_primary":"knowledge_general","capability_primary_share":0.6098,"capability_primary_group_count":14,"capability_non_primary_group_count":29},{"model":"RedPajama-INCITE-7B-Instruct","provider":"Together","country":"US","openness":"open","release_date":"2023-05-05","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":82,"benchmark_group_count":40,"benchmark_coverage_score":7.71,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.51,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.7654,"source_primary_group_share":0.575,"capability_breadth":1.93,"capability_primary":"knowledge_general","capability_primary_share":0.7037,"capability_primary_group_count":20,"capability_non_primary_group_count":20},{"model":"GLM-5.1 (Thinking)","provider":"Zhipu","country":"CN","openness":"open","release_date":"2026-04-07","param_count":744.0,"active_param_count":40.0,"training_compute_flop":null,"num_benchmarks":82,"benchmark_group_count":58,"benchmark_coverage_score":33.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.25,"source_primary":"artificialanalysis.ai","source_primary_share":0.4545,"source_primary_group_share":0.6034,"capability_breadth":3.46,"capability_primary":"code","capability_primary_share":0.4342,"capability_primary_group_count":24,"capability_non_primary_group_count":33},{"model":"GPT-5.4 (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-03-06","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":82,"benchmark_group_count":68,"benchmark_coverage_score":51.0,"source_family_count":33,"source_group_family_count":33,"source_effective_breadth":12.3,"source_primary":"rank.opencompass.org.cn","source_primary_share":0.2692,"source_primary_group_share":0.2059,"capability_breadth":6.6,"capability_primary":"knowledge_general","capability_primary_share":0.2436,"capability_primary_group_count":17,"capability_non_primary_group_count":51},{"model":"Qwen 1.5 32B Chat","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-04-02","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":81,"benchmark_group_count":62,"benchmark_coverage_score":9.24,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":2.61,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.6173,"source_primary_group_share":0.5968,"capability_breadth":2.31,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6296,"capability_primary_group_count":38,"capability_non_primary_group_count":24},{"model":"Gemini 1.5 Pro (001)","provider":"Google","country":"US","openness":"api","release_date":"2024-02-15","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 1.x family retired or superseded by newer Gemini models.","num_benchmarks":81,"benchmark_group_count":67,"benchmark_coverage_score":10.94,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":2.3,"source_primary":"crfm.stanford.edu","source_primary_share":0.6364,"source_primary_group_share":0.6418,"capability_breadth":2.74,"capability_primary":"knowledge_general","capability_primary_share":0.5714,"capability_primary_group_count":41,"capability_non_primary_group_count":26},{"model":"Claude Mythos Preview","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-04-07","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":81,"benchmark_group_count":79,"benchmark_coverage_score":56.0,"source_family_count":12,"source_group_family_count":11,"source_effective_breadth":2.88,"source_primary":"www-cdn.anthropic.com","source_primary_share":0.4691,"source_primary_group_share":0.481,"capability_breadth":5.63,"capability_primary":"knowledge_general","capability_primary_share":0.2963,"capability_primary_group_count":23,"capability_non_primary_group_count":56},{"model":"GPT-5.4 Mini (xHigh)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-03-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":81,"benchmark_group_count":47,"benchmark_coverage_score":34.0,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":4.05,"source_primary":"artificialanalysis.ai","source_primary_share":0.6667,"source_primary_group_share":0.4468,"capability_breadth":5.85,"capability_primary":"code","capability_primary_share":0.2625,"capability_primary_group_count":12,"capability_non_primary_group_count":34},{"model":"granite-3.1-8B-instruct","provider":"IBM","country":"US","openness":"open","release_date":"2024-12-06","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":80,"benchmark_group_count":62,"benchmark_coverage_score":7.06,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":2.5,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.625,"source_primary_group_share":0.5968,"capability_breadth":1.77,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7375,"capability_primary_group_count":44,"capability_non_primary_group_count":18},{"model":"DeepSeek V3.2 (Non-reasoning)","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-12-01","param_count":671.0,"active_param_count":37.0,"training_compute_flop":4.2e+24,"num_benchmarks":80,"benchmark_group_count":48,"benchmark_coverage_score":39.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":3.11,"source_primary":"artificialanalysis.ai","source_primary_share":0.6456,"source_primary_group_share":0.4167,"capability_breadth":6.51,"capability_primary":"knowledge_general","capability_primary_share":0.2152,"capability_primary_group_count":9,"capability_non_primary_group_count":39},{"model":"Seed-OSS-36B-Instruct","provider":"ByteDance","country":"CN","openness":"open","release_date":"2025-10-01","param_count":36.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":79,"benchmark_group_count":44,"benchmark_coverage_score":33.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":3.69,"source_primary":"artificialanalysis.ai","source_primary_share":0.6538,"source_primary_group_share":0.4545,"capability_breadth":4.98,"capability_primary":"knowledge_general","capability_primary_share":0.2821,"capability_primary_group_count":11,"capability_non_primary_group_count":33},{"model":"GPT-4.5 (Preview)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-02-27","param_count":null,"active_param_count":null,"training_compute_flop":3.8e+26,"lifecycle":"deprecated","lifecycle_reason":"OpenAI GPT-4.5 preview model retired or superseded by newer API models.","num_benchmarks":79,"benchmark_group_count":66,"benchmark_coverage_score":47.0,"source_family_count":30,"source_group_family_count":30,"source_effective_breadth":14.52,"source_primary":"fab-ai.org","source_primary_share":0.1818,"source_primary_group_share":0.1667,"capability_breadth":6.62,"capability_primary":"vision_language","capability_primary_share":0.2727,"capability_primary_group_count":19,"capability_non_primary_group_count":47},{"model":"Gemma 1.1 2B (IT)","provider":"Google","country":"US","openness":"open","release_date":"2024-04-09","param_count":2.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":79,"benchmark_group_count":60,"benchmark_coverage_score":8.82,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":2.45,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.6329,"source_primary_group_share":0.6167,"capability_breadth":2.2,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6456,"capability_primary_group_count":38,"capability_non_primary_group_count":22},{"model":"Nova Lite","provider":"Amazon","country":"US","openness":"api","release_date":"2024-12-03","param_count":20.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":79,"benchmark_group_count":54,"benchmark_coverage_score":41.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":4.46,"source_primary":"artificialanalysis.ai","source_primary_share":0.4557,"source_primary_group_share":0.3333,"capability_breadth":5.03,"capability_primary":"knowledge_general","capability_primary_share":0.3418,"capability_primary_group_count":13,"capability_non_primary_group_count":41},{"model":"opt-30B","provider":"Meta","country":"US","openness":"open","release_date":"2022-05-11","param_count":30.0,"active_param_count":null,"training_compute_flop":5.4e+22,"num_benchmarks":79,"benchmark_group_count":37,"benchmark_coverage_score":8.62,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.23,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.7848,"source_primary_group_share":0.6216,"capability_breadth":2.16,"capability_primary":"knowledge_general","capability_primary_share":0.6582,"capability_primary_group_count":16,"capability_non_primary_group_count":21},{"model":"StableLM Tuned Alpha 7B","provider":"Stability AI","country":"UK","openness":"open","release_date":"2023-04-25","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":79,"benchmark_group_count":37,"benchmark_coverage_score":8.07,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.23,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.7848,"source_primary_group_share":0.6216,"capability_breadth":2.02,"capability_primary":"knowledge_general","capability_primary_share":0.6835,"capability_primary_group_count":18,"capability_non_primary_group_count":19},{"model":"Kimi K2.5 (Non-reasoning)","provider":"Moonshot","country":"CN","openness":"open","release_date":"2026-01-27","param_count":1000.0,"active_param_count":32.0,"training_compute_flop":5.8e+24,"num_benchmarks":79,"benchmark_group_count":46,"benchmark_coverage_score":34.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.13,"source_primary":"artificialanalysis.ai","source_primary_share":0.6282,"source_primary_group_share":0.3913,"capability_breadth":6.57,"capability_primary":"knowledge_general","capability_primary_share":0.2051,"capability_primary_group_count":12,"capability_non_primary_group_count":34},{"model":"Qwen3 Coder Next","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-02-03","param_count":80.0,"active_param_count":3.0,"training_compute_flop":null,"num_benchmarks":79,"benchmark_group_count":45,"benchmark_coverage_score":32.0,"source_family_count":14,"source_group_family_count":12,"source_effective_breadth":4.23,"source_primary":"artificialanalysis.ai","source_primary_share":0.6456,"source_primary_group_share":0.4222,"capability_breadth":4.78,"capability_primary":"code","capability_primary_share":0.2911,"capability_primary_group_count":13,"capability_non_primary_group_count":32},{"model":"Gemini 1.5 Flash-8B","provider":"Google","country":"US","openness":"api","release_date":"2024-10-03","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 1.x family retired or superseded by newer Gemini models.","num_benchmarks":79,"benchmark_group_count":57,"benchmark_coverage_score":38.0,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":4.76,"source_primary":"huggingface.co/spaces/av-odyssey/av_odyssey_bench_leaderboard","source_primary_share":0.3235,"source_primary_group_share":0.3684,"capability_breadth":6.15,"capability_primary":"vision_language","capability_primary_share":0.2941,"capability_primary_group_count":19,"capability_non_primary_group_count":38},{"model":"Llama 4 Maverick Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2025-04-05","param_count":400.0,"active_param_count":17.0,"training_compute_flop":2.244e+24,"num_benchmarks":78,"benchmark_group_count":73,"benchmark_coverage_score":54.0,"source_family_count":21,"source_group_family_count":21,"source_effective_breadth":8.87,"source_primary":"vals.ai","source_primary_share":0.2208,"source_primary_group_share":0.2329,"capability_breadth":6.07,"capability_primary":"knowledge_general","capability_primary_share":0.2597,"capability_primary_group_count":19,"capability_non_primary_group_count":54},{"model":"GPT-NeoXT-Chat-Base-20B","provider":"Together","country":"US","openness":"open","release_date":"2023-03-03","param_count":20.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":78,"benchmark_group_count":36,"benchmark_coverage_score":8.41,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.11,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.7949,"source_primary_group_share":0.6389,"capability_breadth":2.1,"capability_primary":"knowledge_general","capability_primary_share":0.6667,"capability_primary_group_count":16,"capability_non_primary_group_count":20},{"model":"Qwen2.5-14B-Instruct-1M","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-01-27","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":78,"benchmark_group_count":65,"benchmark_coverage_score":8.61,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":2.68,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.641,"source_primary_group_share":0.5692,"capability_breadth":2.15,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6667,"capability_primary_group_count":42,"capability_non_primary_group_count":23},{"model":"Qwen 2 57B A14B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-06-04","param_count":57.0,"active_param_count":14.0,"training_compute_flop":3.78e+23,"num_benchmarks":78,"benchmark_group_count":59,"benchmark_coverage_score":7.19,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.31,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.641,"source_primary_group_share":0.6271,"capability_breadth":1.8,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7308,"capability_primary_group_count":42,"capability_non_primary_group_count":17},{"model":"bloom-7b1","provider":"Other","country":"XX","openness":"open","release_date":"2022-05-19","param_count":7.1,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":78,"benchmark_group_count":33,"benchmark_coverage_score":8.41,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.89,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.7949,"source_primary_group_share":0.697,"capability_breadth":2.1,"capability_primary":"knowledge_general","capability_primary_share":0.6667,"capability_primary_group_count":16,"capability_non_primary_group_count":17},{"model":"StableBeluga2","provider":"Stability AI","country":"UK","openness":"open","release_date":"2023-07-27","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":78,"benchmark_group_count":38,"benchmark_coverage_score":7.66,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.29,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.7949,"source_primary_group_share":0.6053,"capability_breadth":1.91,"capability_primary":"knowledge_general","capability_primary_share":0.7051,"capability_primary_group_count":19,"capability_non_primary_group_count":19},{"model":"O1 Preview","provider":"OpenAI","country":"US","openness":"api","release_date":"2024-09-12","param_count":200.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI preview model superseded by later o-series models.","num_benchmarks":77,"benchmark_group_count":66,"benchmark_coverage_score":47.0,"source_family_count":34,"source_group_family_count":34,"source_effective_breadth":15.12,"source_primary":"benchmarklist.com","source_primary_share":0.1408,"source_primary_group_share":0.1515,"capability_breadth":6.75,"capability_primary":"code","capability_primary_share":0.2676,"capability_primary_group_count":19,"capability_non_primary_group_count":47},{"model":"Command-R+","provider":"Cohere","country":"CA","openness":"open","release_date":"2024-04-04","param_count":104.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":77,"benchmark_group_count":69,"benchmark_coverage_score":51.0,"source_family_count":27,"source_group_family_count":27,"source_effective_breadth":9.1,"source_primary":"livebench.ai","source_primary_share":0.2208,"source_primary_group_share":0.2464,"capability_breadth":6.71,"capability_primary":"knowledge_general","capability_primary_share":0.2727,"capability_primary_group_count":18,"capability_non_primary_group_count":51},{"model":"CodeLlama-34B-hf","provider":"Meta","country":"US","openness":"open","release_date":"2023-08-24","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":77,"benchmark_group_count":33,"benchmark_coverage_score":8.45,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.87,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8267,"source_primary_group_share":0.697,"capability_breadth":2.11,"capability_primary":"knowledge_general","capability_primary_share":0.6667,"capability_primary_group_count":14,"capability_non_primary_group_count":19},{"model":"CodeLlama-13B-Instruct-hf","provider":"Meta","country":"US","openness":"open","release_date":"2023-08-24","param_count":13.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":77,"benchmark_group_count":33,"benchmark_coverage_score":8.43,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.87,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8267,"source_primary_group_share":0.697,"capability_breadth":2.11,"capability_primary":"knowledge_general","capability_primary_share":0.6667,"capability_primary_group_count":14,"capability_non_primary_group_count":19},{"model":"CodeLlama-13B-hf","provider":"Meta","country":"US","openness":"open","release_date":"2023-08-24","param_count":13.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":77,"benchmark_group_count":33,"benchmark_coverage_score":8.43,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.87,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8267,"source_primary_group_share":0.697,"capability_breadth":2.11,"capability_primary":"knowledge_general","capability_primary_share":0.6667,"capability_primary_group_count":14,"capability_non_primary_group_count":19},{"model":"CodeLlama-7B-Instruct-hf","provider":"Meta","country":"US","openness":"open","release_date":"2023-08-24","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":77,"benchmark_group_count":33,"benchmark_coverage_score":8.43,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.87,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8267,"source_primary_group_share":0.697,"capability_breadth":2.11,"capability_primary":"knowledge_general","capability_primary_share":0.6667,"capability_primary_group_count":14,"capability_non_primary_group_count":19},{"model":"CodeLlama-7B-hf","provider":"Meta","country":"US","openness":"open","release_date":"2023-08-24","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":77,"benchmark_group_count":33,"benchmark_coverage_score":8.43,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.87,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8267,"source_primary_group_share":0.697,"capability_breadth":2.11,"capability_primary":"knowledge_general","capability_primary_share":0.6667,"capability_primary_group_count":14,"capability_non_primary_group_count":19},{"model":"DeepSeek V4 Pro (Reasoning, Max Effort)","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2026-04-23","param_count":1600.0,"active_param_count":49.0,"training_compute_flop":9.702e+24,"num_benchmarks":77,"benchmark_group_count":54,"benchmark_coverage_score":12.78,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.76,"source_primary":"qwen.ai","source_primary_share":0.5068,"source_primary_group_share":0.6852,"capability_breadth":3.2,"capability_primary":"code","capability_primary_share":0.4722,"capability_primary_group_count":25,"capability_non_primary_group_count":28},{"model":"trurl-2-7B","provider":"VoiceLab","country":"PL","openness":"open","release_date":"2023-08-16","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":76,"benchmark_group_count":34,"benchmark_coverage_score":8.26,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.89,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8158,"source_primary_group_share":0.6765,"capability_breadth":2.07,"capability_primary":"knowledge_general","capability_primary_share":0.6711,"capability_primary_group_count":15,"capability_non_primary_group_count":19},{"model":"Qwen 2 1.5B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-06-06","param_count":1.5,"active_param_count":null,"training_compute_flop":6.3e+22,"num_benchmarks":76,"benchmark_group_count":54,"benchmark_coverage_score":7.48,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.0,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.6579,"source_primary_group_share":0.6852,"capability_breadth":1.87,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7105,"capability_primary_group_count":40,"capability_non_primary_group_count":14},{"model":"GPT-4 (0613)","provider":"OpenAI","country":"US","openness":"api","release_date":"2023-06-13","param_count":1760.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI GPT-4/GPT-4 Turbo family retired or superseded by newer API models.","num_benchmarks":75,"benchmark_group_count":70,"benchmark_coverage_score":59.0,"source_family_count":25,"source_group_family_count":25,"source_effective_breadth":9.68,"source_primary":"livebench.ai","source_primary_share":0.2267,"source_primary_group_share":0.2429,"capability_breadth":7.57,"capability_primary":"knowledge_multilingual","capability_primary_share":0.2,"capability_primary_group_count":11,"capability_non_primary_group_count":59},{"model":"mpt-7B-chat","provider":"Databricks","country":"US","openness":"open","release_date":"2023-05-22","param_count":7.0,"active_param_count":null,"training_compute_flop":4.2e+22,"num_benchmarks":75,"benchmark_group_count":35,"benchmark_coverage_score":7.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.05,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8378,"source_primary_group_share":0.6571,"capability_breadth":1.75,"capability_primary":"knowledge_general","capability_primary_share":0.7432,"capability_primary_group_count":19,"capability_non_primary_group_count":16},{"model":"mpt-30B Instruct","provider":"Databricks","country":"US","openness":"open","release_date":"2023-06-22","param_count":30.0,"active_param_count":null,"training_compute_flop":1.89e+23,"num_benchmarks":75,"benchmark_group_count":34,"benchmark_coverage_score":8.15,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.92,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8267,"source_primary_group_share":0.6765,"capability_breadth":2.04,"capability_primary":"knowledge_general","capability_primary_share":0.68,"capability_primary_group_count":15,"capability_non_primary_group_count":19},{"model":"MythoMax-L2-13B","provider":"Other","country":"XX","openness":"open","release_date":null,"param_count":13.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":75,"benchmark_group_count":29,"benchmark_coverage_score":8.53,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.52,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8267,"source_primary_group_share":0.7931,"capability_breadth":2.13,"capability_primary":"knowledge_general","capability_primary_share":0.6667,"capability_primary_group_count":14,"capability_non_primary_group_count":15},{"model":"Nous-Hermes-llama-2-7B","provider":"Nous Research","country":"US","openness":"open","release_date":"2023-07-25","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":75,"benchmark_group_count":35,"benchmark_coverage_score":7.54,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.07,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8267,"source_primary_group_share":0.6571,"capability_breadth":1.88,"capability_primary":"knowledge_general","capability_primary_share":0.7067,"capability_primary_group_count":17,"capability_non_primary_group_count":18},{"model":"GPT-5 Nano (Medium)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-08-07","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":74,"benchmark_group_count":43,"benchmark_coverage_score":33.0,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":4.46,"source_primary":"artificialanalysis.ai","source_primary_share":0.473,"source_primary_group_share":0.3953,"capability_breadth":5.49,"capability_primary":"knowledge_general","capability_primary_share":0.2568,"capability_primary_group_count":10,"capability_non_primary_group_count":33},{"model":"vicuna-13B","provider":"LMSYS","country":"US","openness":"open","release_date":"2023-04-03","param_count":13.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":74,"benchmark_group_count":33,"benchmark_coverage_score":7.83,"source_family_count":11,"source_group_family_count":10,"source_effective_breadth":2.01,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8378,"source_primary_group_share":0.697,"capability_breadth":1.96,"capability_primary":"knowledge_general","capability_primary_share":0.7027,"capability_primary_group_count":16,"capability_non_primary_group_count":17},{"model":"falcon-180B","provider":"TII","country":"UAE","openness":"open","release_date":"2023-09-06","param_count":180.0,"active_param_count":null,"training_compute_flop":3.76e+24,"num_benchmarks":74,"benchmark_group_count":35,"benchmark_coverage_score":7.24,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.94,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8378,"source_primary_group_share":0.6571,"capability_breadth":1.81,"capability_primary":"knowledge_general","capability_primary_share":0.7297,"capability_primary_group_count":18,"capability_non_primary_group_count":17},{"model":"vicuna-33B-v1.3","provider":"LMSYS","country":"US","openness":"open","release_date":"2023-06-21","param_count":33.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":74,"benchmark_group_count":33,"benchmark_coverage_score":8.05,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":1.97,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8378,"source_primary_group_share":0.697,"capability_breadth":2.01,"capability_primary":"knowledge_general","capability_primary_share":0.6892,"capability_primary_group_count":15,"capability_non_primary_group_count":18},{"model":"vicuna-13B-v1.5","provider":"LMSYS","country":"US","openness":"open","release_date":"2023-07-29","param_count":13.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":74,"benchmark_group_count":30,"benchmark_coverage_score":7.99,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":1.66,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8378,"source_primary_group_share":0.7667,"capability_breadth":2.0,"capability_primary":"knowledge_general","capability_primary_share":0.6892,"capability_primary_group_count":15,"capability_non_primary_group_count":15},{"model":"Baichuan-7B","provider":"Baichuan","country":"CN","openness":"open","release_date":"2023-06-15","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":74,"benchmark_group_count":32,"benchmark_coverage_score":7.61,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.84,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8493,"source_primary_group_share":0.7188,"capability_breadth":1.9,"capability_primary":"knowledge_general","capability_primary_share":0.7123,"capability_primary_group_count":16,"capability_non_primary_group_count":16},{"model":"dolly-v2-3B","provider":"Databricks","country":"US","openness":"open","release_date":"2023-04-17","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":74,"benchmark_group_count":34,"benchmark_coverage_score":7.34,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.96,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8378,"source_primary_group_share":0.6765,"capability_breadth":1.84,"capability_primary":"knowledge_general","capability_primary_share":0.7162,"capability_primary_group_count":17,"capability_non_primary_group_count":17},{"model":"CodeLlama-34B-Python-hf","provider":"Meta","country":"US","openness":"open","release_date":"2023-08-24","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":73,"benchmark_group_count":31,"benchmark_coverage_score":8.02,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.66,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8493,"source_primary_group_share":0.7419,"capability_breadth":2.01,"capability_primary":"knowledge_general","capability_primary_share":0.6849,"capability_primary_group_count":14,"capability_non_primary_group_count":17},{"model":"Starling-LM-7B-beta","provider":"UC Berkeley","country":"US","openness":"open","release_date":"2023-11-30","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":73,"benchmark_group_count":47,"benchmark_coverage_score":30.0,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":3.86,"source_primary":"euroeval.com","source_primary_share":0.4247,"source_primary_group_share":0.3617,"capability_breadth":4.02,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3699,"capability_primary_group_count":17,"capability_non_primary_group_count":30},{"model":"OpenHermes-13B","provider":"Other","country":"XX","openness":"open","release_date":"2023-09-06","param_count":13.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":73,"benchmark_group_count":31,"benchmark_coverage_score":7.56,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.69,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8493,"source_primary_group_share":0.7419,"capability_breadth":1.89,"capability_primary":"knowledge_general","capability_primary_share":0.7123,"capability_primary_group_count":16,"capability_non_primary_group_count":15},{"model":"RedPajama-INCITE-7B-Chat","provider":"Together","country":"US","openness":"open","release_date":"2023-05-04","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":73,"benchmark_group_count":32,"benchmark_coverage_score":7.05,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.78,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8493,"source_primary_group_share":0.7188,"capability_breadth":1.76,"capability_primary":"knowledge_general","capability_primary_share":0.7397,"capability_primary_group_count":17,"capability_non_primary_group_count":15},{"model":"RedPajama-INCITE-Instruct-3B-v1","provider":"Together","country":"US","openness":"open","release_date":"2023-05-05","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":73,"benchmark_group_count":32,"benchmark_coverage_score":7.34,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.78,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8493,"source_primary_group_share":0.7188,"capability_breadth":1.83,"capability_primary":"knowledge_general","capability_primary_share":0.726,"capability_primary_group_count":17,"capability_non_primary_group_count":15},{"model":"RedPajama-INCITE-Base-3B-v1","provider":"Together","country":"US","openness":"open","release_date":"2023-05-04","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":73,"benchmark_group_count":32,"benchmark_coverage_score":7.34,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.78,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8493,"source_primary_group_share":0.7188,"capability_breadth":1.83,"capability_primary":"knowledge_general","capability_primary_share":0.726,"capability_primary_group_count":17,"capability_non_primary_group_count":15},{"model":"bloom-1b7","provider":"Other","country":"XX","openness":"open","release_date":"2022-05-19","param_count":1.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":73,"benchmark_group_count":30,"benchmark_coverage_score":7.56,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.59,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8493,"source_primary_group_share":0.7667,"capability_breadth":1.89,"capability_primary":"knowledge_general","capability_primary_share":0.7123,"capability_primary_group_count":16,"capability_non_primary_group_count":14},{"model":"OLMo 3 32B (Thinking)","provider":"Allen AI","country":"US","openness":"open","release_date":"2025-11-20","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":73,"benchmark_group_count":49,"benchmark_coverage_score":11.07,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":3.2,"source_primary":"artificialanalysis.ai","source_primary_share":0.4722,"source_primary_group_share":0.4286,"capability_breadth":2.77,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":26,"capability_non_primary_group_count":23},{"model":"Qwen 3 14B (Reasoning)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":14.0,"active_param_count":null,"training_compute_flop":3.197e+24,"num_benchmarks":73,"benchmark_group_count":44,"benchmark_coverage_score":9.74,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.38,"source_primary":"artificialanalysis.ai","source_primary_share":0.5278,"source_primary_group_share":0.4773,"capability_breadth":2.44,"capability_primary":"knowledge_general","capability_primary_share":0.5972,"capability_primary_group_count":26,"capability_non_primary_group_count":18},{"model":"O3 (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-04-16","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":72,"benchmark_group_count":66,"benchmark_coverage_score":52.0,"source_family_count":28,"source_group_family_count":28,"source_effective_breadth":11.06,"source_primary":"cdn.openai.com","source_primary_share":0.2319,"source_primary_group_share":0.2424,"capability_breadth":5.97,"capability_primary":"math","capability_primary_share":0.2029,"capability_primary_group_count":14,"capability_non_primary_group_count":52},{"model":"LongChat-7B-v1.5-32k","provider":"Other","country":"XX","openness":"open","release_date":"2023-08-01","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":72,"benchmark_group_count":33,"benchmark_coverage_score":7.82,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.78,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8611,"source_primary_group_share":0.697,"capability_breadth":1.96,"capability_primary":"knowledge_general","capability_primary_share":0.6944,"capability_primary_group_count":14,"capability_non_primary_group_count":19},{"model":"Human Expert","provider":"Human","country":null,"openness":null,"release_date":null,"param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":72,"benchmark_group_count":59,"benchmark_coverage_score":37.0,"source_family_count":33,"source_group_family_count":33,"source_effective_breadth":12.75,"source_primary":"matharena.ai","source_primary_share":0.2727,"source_primary_group_share":0.2203,"capability_breadth":3.31,"capability_primary":"math","capability_primary_share":0.4091,"capability_primary_group_count":22,"capability_non_primary_group_count":37},{"model":"Llama 2 13B","provider":"Meta","country":"US","openness":"open","release_date":"2023-07-18","param_count":13.0,"active_param_count":null,"training_compute_flop":1.6e+23,"num_benchmarks":72,"benchmark_group_count":65,"benchmark_coverage_score":39.0,"source_family_count":15,"source_group_family_count":15,"source_effective_breadth":4.38,"source_primary":"crfm.stanford.edu","source_primary_share":0.375,"source_primary_group_share":0.4154,"capability_breadth":4.98,"capability_primary":"knowledge_general","capability_primary_share":0.375,"capability_primary_group_count":26,"capability_non_primary_group_count":39},{"model":"CodeLlama-13B-Python-hf","provider":"Meta","country":"US","openness":"open","release_date":"2023-08-24","param_count":13.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":72,"benchmark_group_count":30,"benchmark_coverage_score":7.81,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.56,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8611,"source_primary_group_share":0.7667,"capability_breadth":1.95,"capability_primary":"knowledge_general","capability_primary_share":0.6944,"capability_primary_group_count":14,"capability_non_primary_group_count":16},{"model":"CodeLlama-7B-Python-hf","provider":"Meta","country":"US","openness":"open","release_date":"2023-08-24","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":72,"benchmark_group_count":30,"benchmark_coverage_score":7.81,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.56,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8611,"source_primary_group_share":0.7667,"capability_breadth":1.95,"capability_primary":"knowledge_general","capability_primary_share":0.6944,"capability_primary_group_count":14,"capability_non_primary_group_count":16},{"model":"codegen-16B-nl","provider":"Salesforce","country":"US","openness":"open","release_date":"2022-04-13","param_count":16.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":72,"benchmark_group_count":30,"benchmark_coverage_score":7.81,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.56,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8611,"source_primary_group_share":0.7667,"capability_breadth":1.95,"capability_primary":"knowledge_general","capability_primary_share":0.6944,"capability_primary_group_count":14,"capability_non_primary_group_count":16},{"model":"stablelm-base-alpha-7B","provider":"Stability AI","country":"UK","openness":"open","release_date":"2023-04-11","param_count":7.0,"active_param_count":null,"training_compute_flop":4.5e+22,"num_benchmarks":72,"benchmark_group_count":30,"benchmark_coverage_score":7.81,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.56,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8611,"source_primary_group_share":0.7667,"capability_breadth":1.95,"capability_primary":"knowledge_general","capability_primary_share":0.6944,"capability_primary_group_count":14,"capability_non_primary_group_count":16},{"model":"stablelm-tuned-alpha-3B","provider":"Stability AI","country":"UK","openness":"open","release_date":"2023-04-19","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":72,"benchmark_group_count":30,"benchmark_coverage_score":7.81,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.56,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8611,"source_primary_group_share":0.7667,"capability_breadth":1.95,"capability_primary":"knowledge_general","capability_primary_share":0.6944,"capability_primary_group_count":14,"capability_non_primary_group_count":16},{"model":"Qwen2-Audio-7B-Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-07-31","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":72,"benchmark_group_count":56,"benchmark_coverage_score":36.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.4,"source_primary":"github.com/audiollms/audiobench","source_primary_share":0.8361,"source_primary_group_share":0.8393,"capability_breadth":4.06,"capability_primary":"structured_precision","capability_primary_share":0.3934,"capability_primary_group_count":20,"capability_non_primary_group_count":36},{"model":"Grok 4.1 Fast (Non-reasoning)","provider":"xAI","country":"US","openness":"api","release_date":"2025-11-19","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":72,"benchmark_group_count":50,"benchmark_coverage_score":36.0,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":4.07,"source_primary":"artificialanalysis.ai","source_primary_share":0.5,"source_primary_group_share":0.34,"capability_breadth":4.93,"capability_primary":"knowledge_general","capability_primary_share":0.3143,"capability_primary_group_count":14,"capability_non_primary_group_count":36},{"model":"qwen3.5-flash","provider":"Alibaba","country":"CN","openness":"api","release_date":"2026-02-16","param_count":35.0,"active_param_count":3.0,"training_compute_flop":null,"num_benchmarks":72,"benchmark_group_count":57,"benchmark_coverage_score":43.0,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":5.63,"source_primary":"fab-ai.org","source_primary_share":0.3,"source_primary_group_share":0.2807,"capability_breadth":5.3,"capability_primary":"knowledge_general","capability_primary_share":0.2857,"capability_primary_group_count":14,"capability_non_primary_group_count":43},{"model":"Gemini 2.0 Flash Lite","provider":"Google","country":"US","openness":"api","release_date":"2025-02-25","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 2.0 family deprecated in favor of newer Gemini models.","num_benchmarks":71,"benchmark_group_count":62,"benchmark_coverage_score":35.0,"source_family_count":18,"source_group_family_count":18,"source_effective_breadth":3.12,"source_primary":"crfm.stanford.edu","source_primary_share":0.493,"source_primary_group_share":0.5484,"capability_breadth":4.7,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3803,"capability_primary_group_count":27,"capability_non_primary_group_count":35},{"model":"Qwen2.5-Math-7B-Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-19","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":71,"benchmark_group_count":54,"benchmark_coverage_score":8.49,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":2.04,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.7042,"source_primary_group_share":0.6852,"capability_breadth":2.12,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6479,"capability_primary_group_count":36,"capability_non_primary_group_count":18},{"model":"WizardLM-13B-V1.2","provider":"Microsoft","country":"US","openness":"open","release_date":"2023-07-25","param_count":13.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":71,"benchmark_group_count":32,"benchmark_coverage_score":7.18,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8732,"source_primary_group_share":0.7188,"capability_breadth":1.79,"capability_primary":"knowledge_general","capability_primary_share":0.7324,"capability_primary_group_count":16,"capability_non_primary_group_count":16},{"model":"Qwen1.5-MoE-A2.7B-Chat","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-03-14","param_count":7.0,"active_param_count":2.7,"training_compute_flop":null,"num_benchmarks":71,"benchmark_group_count":57,"benchmark_coverage_score":8.77,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.18,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.7042,"source_primary_group_share":0.6491,"capability_breadth":2.19,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6479,"capability_primary_group_count":36,"capability_non_primary_group_count":21},{"model":"GPT-5.2 (Non-reasoning)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-12-11","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":71,"benchmark_group_count":33,"benchmark_coverage_score":21.3,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":2.36,"source_primary":"artificialanalysis.ai","source_primary_share":0.7571,"source_primary_group_share":0.6364,"capability_breadth":5.33,"capability_primary":"knowledge_general","capability_primary_share":0.2571,"capability_primary_group_count":8,"capability_non_primary_group_count":25},{"model":"Grok 4 Fast (Non-reasoning)","provider":"xAI","country":"US","openness":"api","release_date":"2025-09-19","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":71,"benchmark_group_count":49,"benchmark_coverage_score":35.0,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":4.33,"source_primary":"artificialanalysis.ai","source_primary_share":0.5072,"source_primary_group_share":0.3469,"capability_breadth":4.88,"capability_primary":"knowledge_general","capability_primary_share":0.3188,"capability_primary_group_count":14,"capability_non_primary_group_count":35},{"model":"GLM-5 (Non-reasoning)","provider":"Zhipu","country":"CN","openness":"open","release_date":"2026-02-12","param_count":744.0,"active_param_count":40.0,"training_compute_flop":6.84e+24,"num_benchmarks":71,"benchmark_group_count":40,"benchmark_coverage_score":32.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.43,"source_primary":"artificialanalysis.ai","source_primary_share":0.6857,"source_primary_group_share":0.45,"capability_breadth":6.3,"capability_primary":"knowledge_general","capability_primary_share":0.2286,"capability_primary_group_count":8,"capability_non_primary_group_count":32},{"model":"GLM-4.7 FP8","provider":"Zhipu","country":"CN","openness":"open","release_date":"2025-12-22","param_count":355.0,"active_param_count":32.0,"training_compute_flop":4.42e+24,"num_benchmarks":71,"benchmark_group_count":55,"benchmark_coverage_score":9.89,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.47,"source_primary":"medmarks.ai","source_primary_share":0.6056,"source_primary_group_share":0.6,"capability_breadth":2.47,"capability_primary":"knowledge_general","capability_primary_share":0.6056,"capability_primary_group_count":32,"capability_non_primary_group_count":23},{"model":"Qwen 2.5 VL 7B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-01-28","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":70,"benchmark_group_count":59,"benchmark_coverage_score":8.49,"source_family_count":14,"source_group_family_count":14,"source_effective_breadth":5.81,"source_primary":"llm-stats.com","source_primary_share":0.2754,"source_primary_group_share":0.2373,"capability_breadth":2.12,"capability_primary":"vision_language","capability_primary_share":0.6667,"capability_primary_group_count":41,"capability_non_primary_group_count":18},{"model":"WizardCoder-Python-34B-V1.0","provider":"Other","country":"XX","openness":"open","release_date":"2023-08-26","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":70,"benchmark_group_count":31,"benchmark_coverage_score":7.41,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":1.75,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8857,"source_primary_group_share":0.7419,"capability_breadth":1.85,"capability_primary":"knowledge_general","capability_primary_share":0.7143,"capability_primary_group_count":14,"capability_non_primary_group_count":17},{"model":"Llama 2 7B 32K","provider":"Meta","country":"US","openness":"open","release_date":"2023-07-18","param_count":7.0,"active_param_count":null,"training_compute_flop":8.4e+22,"num_benchmarks":70,"benchmark_group_count":31,"benchmark_coverage_score":7.01,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.69,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8857,"source_primary_group_share":0.7419,"capability_breadth":1.75,"capability_primary":"knowledge_general","capability_primary_share":0.7429,"capability_primary_group_count":16,"capability_non_primary_group_count":15},{"model":"SALMONN-7B","provider":"Other","country":"XX","openness":"open","release_date":"2023-11-13","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":70,"benchmark_group_count":54,"benchmark_coverage_score":34.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.3,"source_primary":"github.com/audiollms/audiobench","source_primary_share":0.8644,"source_primary_group_share":0.8704,"capability_breadth":3.81,"capability_primary":"structured_precision","capability_primary_share":0.4068,"capability_primary_group_count":20,"capability_non_primary_group_count":34},{"model":"Step 3.7 Flash","provider":"StepFun","country":"CN","openness":"api","release_date":"2026-05-29","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":70,"benchmark_group_count":44,"benchmark_coverage_score":31.0,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":4.7,"source_primary":"artificialanalysis.ai","source_primary_share":0.5143,"source_primary_group_share":0.3636,"capability_breadth":5.92,"capability_primary":"code","capability_primary_share":0.2571,"capability_primary_group_count":13,"capability_non_primary_group_count":31},{"model":"DeepSeek V3.2 Exp (Thinking)","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-09-29","param_count":671.0,"active_param_count":37.0,"training_compute_flop":4.18e+24,"num_benchmarks":70,"benchmark_group_count":36,"benchmark_coverage_score":19.1,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":2.92,"source_primary":"artificialanalysis.ai","source_primary_share":0.7391,"source_primary_group_share":0.5556,"capability_breadth":4.78,"capability_primary":"knowledge_general","capability_primary_share":0.2754,"capability_primary_group_count":9,"capability_non_primary_group_count":27},{"model":"internlm-20B Chat","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2023-09-20","param_count":20.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":69,"benchmark_group_count":30,"benchmark_coverage_score":7.07,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.64,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8986,"source_primary_group_share":0.7667,"capability_breadth":1.77,"capability_primary":"knowledge_general","capability_primary_share":0.7391,"capability_primary_group_count":15,"capability_non_primary_group_count":15},{"model":"Llama 2 70B Chat GPTQ","provider":"Meta","country":"US","openness":"open","release_date":"2023-07-18","param_count":70.0,"active_param_count":null,"training_compute_flop":8.1e+23,"num_benchmarks":69,"benchmark_group_count":30,"benchmark_coverage_score":7.1,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.56,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8986,"source_primary_group_share":0.7667,"capability_breadth":1.77,"capability_primary":"knowledge_general","capability_primary_share":0.7391,"capability_primary_group_count":15,"capability_non_primary_group_count":15},{"model":"trurl-2-13B-academic","provider":"VoiceLab","country":"PL","openness":"open","release_date":"2023-09-18","param_count":13.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":69,"benchmark_group_count":27,"benchmark_coverage_score":7.31,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.37,"source_primary":"huggingface.co/spaces/coreymorris/mmlu-by-task-leaderboard","source_primary_share":0.8986,"source_primary_group_share":0.8519,"capability_breadth":1.83,"capability_primary":"knowledge_general","capability_primary_share":0.7246,"capability_primary_group_count":14,"capability_non_primary_group_count":13},{"model":"Hy3","provider":"Tencent","country":"CN","openness":"open","release_date":"2026-07-06","param_count":295.0,"active_param_count":21.0,"training_compute_flop":null,"num_benchmarks":69,"benchmark_group_count":47,"benchmark_coverage_score":32.0,"source_family_count":15,"source_group_family_count":15,"source_effective_breadth":5.56,"source_primary":"artificialanalysis.ai","source_primary_share":0.4627,"source_primary_group_share":0.2766,"capability_breadth":4.51,"capability_primary":"code","capability_primary_share":0.3433,"capability_primary_group_count":15,"capability_non_primary_group_count":32},{"model":"Ring-1T","provider":"InclusionAI","country":"CN","openness":"open","release_date":"2025-07-01","param_count":1000.0,"active_param_count":50.0,"training_compute_flop":null,"num_benchmarks":68,"benchmark_group_count":34,"benchmark_coverage_score":21.98,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":2.66,"source_primary":"artificialanalysis.ai","source_primary_share":0.7612,"source_primary_group_share":0.5882,"capability_breadth":5.49,"capability_primary":"knowledge_general","capability_primary_share":0.2537,"capability_primary_group_count":7,"capability_non_primary_group_count":27},{"model":"SOLAR-10.7B-Instruct-v1.0","provider":"Upstage","country":"KR","openness":"open","release_date":"2023-12-13","param_count":10.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":68,"benchmark_group_count":50,"benchmark_coverage_score":36.0,"source_family_count":14,"source_group_family_count":14,"source_effective_breadth":5.08,"source_primary":"huggingface.co","source_primary_share":0.3433,"source_primary_group_share":0.38,"capability_breadth":4.08,"capability_primary":"knowledge_general","capability_primary_share":0.3433,"capability_primary_group_count":14,"capability_non_primary_group_count":36},{"model":"SmolLM2-1.7B","provider":"HuggingFace","country":"FR","openness":"open","release_date":"2024-10-30","param_count":1.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":68,"benchmark_group_count":40,"benchmark_coverage_score":11.26,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":3.12,"source_primary":"euroeval.com","source_primary_share":0.5441,"source_primary_group_share":0.5,"capability_breadth":2.82,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5,"capability_primary_group_count":21,"capability_non_primary_group_count":19},{"model":"Qwen 2 0.5B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-06-06","param_count":0.5,"active_param_count":null,"training_compute_flop":3.6e+22,"num_benchmarks":68,"benchmark_group_count":51,"benchmark_coverage_score":7.06,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.79,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.7353,"source_primary_group_share":0.7255,"capability_breadth":1.77,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7353,"capability_primary_group_count":37,"capability_non_primary_group_count":14},{"model":"GLM-5 (Thinking)","provider":"Zhipu","country":"CN","openness":"open","release_date":"2026-02-12","param_count":744.0,"active_param_count":40.0,"training_compute_flop":6.84e+24,"num_benchmarks":68,"benchmark_group_count":50,"benchmark_coverage_score":33.0,"source_family_count":16,"source_group_family_count":16,"source_effective_breadth":5.32,"source_primary":"artificialanalysis.ai","source_primary_share":0.4776,"source_primary_group_share":0.3,"capability_breadth":3.93,"capability_primary":"knowledge_general","capability_primary_share":0.3731,"capability_primary_group_count":17,"capability_non_primary_group_count":33},{"model":"Grok 4 Fast","provider":"xAI","country":"US","openness":"api","release_date":"2025-09-19","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":68,"benchmark_group_count":58,"benchmark_coverage_score":45.0,"source_family_count":28,"source_group_family_count":28,"source_effective_breadth":11.14,"source_primary":"benchmarklist.com","source_primary_share":0.197,"source_primary_group_share":0.2241,"capability_breadth":6.76,"capability_primary":"knowledge_general","capability_primary_share":0.2879,"capability_primary_group_count":13,"capability_non_primary_group_count":45},{"model":"Mistral 7B Instruct","provider":"Mistral","country":"FR","openness":"open","release_date":"2023-09-27","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":67,"benchmark_group_count":42,"benchmark_coverage_score":31.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":4.24,"source_primary":"artificialanalysis.ai","source_primary_share":0.5077,"source_primary_group_share":0.4048,"capability_breadth":4.43,"capability_primary":"code","capability_primary_share":0.3692,"capability_primary_group_count":11,"capability_non_primary_group_count":31},{"model":"Muse Spark 1.1","provider":"Meta","country":"US","openness":"private","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":67,"benchmark_group_count":63,"benchmark_coverage_score":47.0,"source_family_count":27,"source_group_family_count":27,"source_effective_breadth":6.56,"source_primary":"vals.ai","source_primary_share":0.3636,"source_primary_group_share":0.3492,"capability_breadth":5.97,"capability_primary":"code","capability_primary_share":0.2576,"capability_primary_group_count":16,"capability_non_primary_group_count":47},{"model":"Phi-3 Mini 128K Instruct","provider":"Microsoft","country":"US","openness":"open","release_date":"2024-04-23","param_count":3.8,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":66,"benchmark_group_count":58,"benchmark_coverage_score":47.0,"source_family_count":17,"source_group_family_count":17,"source_effective_breadth":6.62,"source_primary":"livebench.ai","source_primary_share":0.2576,"source_primary_group_share":0.2931,"capability_breadth":7.26,"capability_primary":"knowledge_general","capability_primary_share":0.2121,"capability_primary_group_count":11,"capability_non_primary_group_count":47},{"model":"Qwen 2.5 VL 32B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-03-24","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":65,"benchmark_group_count":59,"benchmark_coverage_score":11.41,"source_family_count":17,"source_group_family_count":17,"source_effective_breadth":7.33,"source_primary":"llm-stats.com","source_primary_share":0.2381,"source_primary_group_share":0.2203,"capability_breadth":2.85,"capability_primary":"vision_language","capability_primary_share":0.5397,"capability_primary_group_count":30,"capability_non_primary_group_count":29},{"model":"GPT-5.4 Pro (xHigh)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-03-06","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":65,"benchmark_group_count":63,"benchmark_coverage_score":46.0,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":4.73,"source_primary":"openai.com","source_primary_share":0.375,"source_primary_group_share":0.381,"capability_breadth":5.33,"capability_primary":"knowledge_general","capability_primary_share":0.2656,"capability_primary_group_count":17,"capability_non_primary_group_count":46},{"model":"Qwen 2 57B A14B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-06-06","param_count":57.0,"active_param_count":14.0,"training_compute_flop":3.78e+23,"num_benchmarks":65,"benchmark_group_count":51,"benchmark_coverage_score":7.55,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.79,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.7692,"source_primary_group_share":0.7255,"capability_breadth":1.89,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7077,"capability_primary_group_count":36,"capability_non_primary_group_count":15},{"model":"Nova Micro","provider":"Amazon","country":"US","openness":"api","release_date":"2024-12-03","param_count":11.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":65,"benchmark_group_count":39,"benchmark_coverage_score":16.08,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":3.39,"source_primary":"artificialanalysis.ai","source_primary_share":0.5385,"source_primary_group_share":0.4615,"capability_breadth":4.02,"capability_primary":"knowledge_general","capability_primary_share":0.4,"capability_primary_group_count":12,"capability_non_primary_group_count":27},{"model":"Qwen1.5-MoE-A2.7B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-02-29","param_count":7.0,"active_param_count":2.7,"training_compute_flop":null,"num_benchmarks":65,"benchmark_group_count":51,"benchmark_coverage_score":7.49,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.79,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.7692,"source_primary_group_share":0.7255,"capability_breadth":1.87,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7077,"capability_primary_group_count":36,"capability_non_primary_group_count":15},{"model":"GLM-5.2 (Max)","provider":"Zhipu","country":"CN","openness":"open","release_date":"2026-06-13","param_count":744.0,"active_param_count":40.0,"training_compute_flop":null,"num_benchmarks":65,"benchmark_group_count":45,"benchmark_coverage_score":30.0,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":4.68,"source_primary":"artificialanalysis.ai","source_primary_share":0.5625,"source_primary_group_share":0.3778,"capability_breadth":4.46,"capability_primary":"code","capability_primary_share":0.3651,"capability_primary_group_count":14,"capability_non_primary_group_count":30},{"model":"DeepSeek V3.1 (Thinking)","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-08-21","param_count":671.0,"active_param_count":37.0,"training_compute_flop":3.594058e+24,"num_benchmarks":65,"benchmark_group_count":44,"benchmark_coverage_score":14.35,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":4.82,"source_primary":"artificialanalysis.ai","source_primary_share":0.5312,"source_primary_group_share":0.3864,"capability_breadth":3.59,"capability_primary":"code","capability_primary_share":0.4531,"capability_primary_group_count":19,"capability_non_primary_group_count":25},{"model":"Qwen 2.5 Coder 14B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-11-12","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":64,"benchmark_group_count":51,"benchmark_coverage_score":7.4,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":1.84,"source_primary":"huggingface.co/spaces/llm-jp/open-japanese-llm-leaderboard","source_primary_share":0.7812,"source_primary_group_share":0.7255,"capability_breadth":1.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7188,"capability_primary_group_count":36,"capability_non_primary_group_count":15},{"model":"Gemini 2.5 Pro (03-25)","provider":"Google","country":"US","openness":"api","release_date":"2025-03-25","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":64,"benchmark_group_count":57,"benchmark_coverage_score":37.0,"source_family_count":14,"source_group_family_count":14,"source_effective_breadth":7.79,"source_primary":"balrogai.com","source_primary_share":0.1746,"source_primary_group_share":0.193,"capability_breadth":5.85,"capability_primary":"vision_language","capability_primary_share":0.3333,"capability_primary_group_count":20,"capability_non_primary_group_count":37},{"model":"Qwen 3 235B A22B 2507 (Thinking)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-07-01","param_count":235.0,"active_param_count":22.0,"training_compute_flop":4.752e+24,"num_benchmarks":64,"benchmark_group_count":27,"benchmark_coverage_score":21.08,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.46,"source_primary":"artificialanalysis.ai","source_primary_share":0.873,"source_primary_group_share":0.8148,"capability_breadth":5.27,"capability_primary":"knowledge_general","capability_primary_share":0.2381,"capability_primary_group_count":6,"capability_non_primary_group_count":21},{"model":"Grok 4.20 Beta (0309) (Reasoning)","provider":"xAI","country":"US","openness":"api","release_date":"2026-03-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":64,"benchmark_group_count":38,"benchmark_coverage_score":9.1,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.03,"source_primary":"euroeval.com","source_primary_share":0.7937,"source_primary_group_share":0.6842,"capability_breadth":2.27,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6349,"capability_primary_group_count":22,"capability_non_primary_group_count":16},{"model":"GPT-5 (Low)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-08-07","param_count":null,"active_param_count":null,"training_compute_flop":6.6e+25,"num_benchmarks":63,"benchmark_group_count":39,"benchmark_coverage_score":31.0,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":4.08,"source_primary":"artificialanalysis.ai","source_primary_share":0.5806,"source_primary_group_share":0.4615,"capability_breadth":5.48,"capability_primary":"code","capability_primary_share":0.2581,"capability_primary_group_count":8,"capability_non_primary_group_count":31},{"model":"c4ai-command-r-v01","provider":"Cohere","country":"CA","openness":"open","release_date":"2024-03-11","param_count":35.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":63,"benchmark_group_count":37,"benchmark_coverage_score":9.28,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":2.38,"source_primary":"euroeval.com","source_primary_share":0.7302,"source_primary_group_share":0.6216,"capability_breadth":2.32,"capability_primary":"knowledge_multilingual","capability_primary_share":0.619,"capability_primary_group_count":23,"capability_non_primary_group_count":14},{"model":"Teuken-7B-Instruct-v0.4","provider":"OpenGPT-X","country":"DE","openness":"open","release_date":"2024-11-26","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":63,"benchmark_group_count":42,"benchmark_coverage_score":10.41,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.95,"source_primary":"euroeval.com","source_primary_share":0.6032,"source_primary_group_share":0.4762,"capability_breadth":2.6,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5714,"capability_primary_group_count":23,"capability_non_primary_group_count":19},{"model":"GPT-5.6 Sol (xHigh)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":63,"benchmark_group_count":42,"benchmark_coverage_score":30.0,"source_family_count":17,"source_group_family_count":17,"source_effective_breadth":5.73,"source_primary":"artificialanalysis.ai","source_primary_share":0.5645,"source_primary_group_share":0.3571,"capability_breadth":4.76,"capability_primary":"code","capability_primary_share":0.3387,"capability_primary_group_count":12,"capability_non_primary_group_count":30},{"model":"Gemini 3 Pro (Preview) (High)","provider":"Google","country":"US","openness":"api","release_date":"2025-11-18","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 3 Pro preview snapshot retired or superseded by newer Gemini models.","num_benchmarks":63,"benchmark_group_count":30,"benchmark_coverage_score":19.22,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":2.16,"source_primary":"artificialanalysis.ai","source_primary_share":0.8387,"source_primary_group_share":0.6667,"capability_breadth":4.8,"capability_primary":"knowledge_general","capability_primary_share":0.2742,"capability_primary_group_count":8,"capability_non_primary_group_count":22},{"model":"Claude Haiku 4.5 (Thinking)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-10-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":63,"benchmark_group_count":28,"benchmark_coverage_score":20.31,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":1.6,"source_primary":"artificialanalysis.ai","source_primary_share":0.9032,"source_primary_group_share":0.7857,"capability_breadth":5.08,"capability_primary":"knowledge_general","capability_primary_share":0.2623,"capability_primary_group_count":6,"capability_non_primary_group_count":21},{"model":"Yi-1.5-9B Chat","provider":"01.AI","country":"CN","openness":"open","release_date":"2024-05-13","param_count":9.0,"active_param_count":null,"training_compute_flop":1.907e+23,"num_benchmarks":62,"benchmark_group_count":53,"benchmark_coverage_score":34.0,"source_family_count":19,"source_group_family_count":19,"source_effective_breadth":7.87,"source_primary":"huggingface.co","source_primary_share":0.2742,"source_primary_group_share":0.2453,"capability_breadth":3.7,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4032,"capability_primary_group_count":19,"capability_non_primary_group_count":34},{"model":"Yi 6B (Base)","provider":"01.AI","country":"CN","openness":"open","release_date":"2023-11-02","param_count":6.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":62,"benchmark_group_count":48,"benchmark_coverage_score":34.0,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":6.82,"source_primary":"euroeval.com","source_primary_share":0.2581,"source_primary_group_share":0.2708,"capability_breadth":5.65,"capability_primary":"reasoning_logical","capability_primary_share":0.2419,"capability_primary_group_count":14,"capability_non_primary_group_count":34},{"model":"Claude Opus 4 (20250514)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-05-14","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 4.0 family deprecated in favor of newer Claude 4.x models.","num_benchmarks":62,"benchmark_group_count":58,"benchmark_coverage_score":45.0,"source_family_count":22,"source_group_family_count":22,"source_effective_breadth":11.29,"source_primary":"vals.ai","source_primary_share":0.1803,"source_primary_group_share":0.1897,"capability_breadth":7.97,"capability_primary":"knowledge_general","capability_primary_share":0.2131,"capability_primary_group_count":13,"capability_non_primary_group_count":45},{"model":"Claude Opus 4.6 (Adaptive Reasoning, Max Effort)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-02-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":62,"benchmark_group_count":29,"benchmark_coverage_score":18.6,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.01,"source_primary":"artificialanalysis.ai","source_primary_share":0.8333,"source_primary_group_share":0.6552,"capability_breadth":4.65,"capability_primary":"knowledge_general","capability_primary_share":0.3,"capability_primary_group_count":7,"capability_non_primary_group_count":22},{"model":"Kimi K2.6 (Thinking)","provider":"Moonshot","country":"CN","openness":"open","release_date":"2026-04-20","param_count":1000.0,"active_param_count":32.0,"training_compute_flop":null,"num_benchmarks":62,"benchmark_group_count":56,"benchmark_coverage_score":35.0,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":2.14,"source_primary":"qwen.ai","source_primary_share":0.6491,"source_primary_group_share":0.6607,"capability_breadth":4.26,"capability_primary":"code","capability_primary_share":0.3684,"capability_primary_group_count":21,"capability_non_primary_group_count":35},{"model":"Mistral-NeMo-Minitron-8B-Instruct","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-10-02","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":62,"benchmark_group_count":38,"benchmark_coverage_score":9.5,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.55,"source_primary":"euroeval.com","source_primary_share":0.629,"source_primary_group_share":0.5,"capability_breadth":2.38,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5968,"capability_primary_group_count":24,"capability_non_primary_group_count":14},{"model":"Ling-1T","provider":"InclusionAI","country":"CN","openness":"open","release_date":"2025-10-09","param_count":1000.0,"active_param_count":50.0,"training_compute_flop":6e+24,"num_benchmarks":61,"benchmark_group_count":41,"benchmark_coverage_score":17.56,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":4.01,"source_primary":"artificialanalysis.ai","source_primary_share":0.5667,"source_primary_group_share":0.4146,"capability_breadth":4.39,"capability_primary":"code","capability_primary_share":0.3833,"capability_primary_group_count":14,"capability_non_primary_group_count":27},{"model":"Yi 6B (Chat)","provider":"01.AI","country":"CN","openness":"open","release_date":"2023-11-02","param_count":6.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":61,"benchmark_group_count":58,"benchmark_coverage_score":44.0,"source_family_count":16,"source_group_family_count":15,"source_effective_breadth":6.98,"source_primary":"livebench.ai","source_primary_share":0.2787,"source_primary_group_share":0.2931,"capability_breadth":6.45,"capability_primary":"knowledge_general","capability_primary_share":0.2623,"capability_primary_group_count":14,"capability_non_primary_group_count":44},{"model":"Gemini 2.0 Flash (Thinking)","provider":"Google","country":"US","openness":"api","release_date":"2024-12-11","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 2.0 family deprecated in favor of newer Gemini models.","num_benchmarks":61,"benchmark_group_count":56,"benchmark_coverage_score":37.0,"source_family_count":25,"source_group_family_count":25,"source_effective_breadth":11.79,"source_primary":"scale.com","source_primary_share":0.1639,"source_primary_group_share":0.1786,"capability_breadth":5.72,"capability_primary":"vision_language","capability_primary_share":0.3279,"capability_primary_group_count":19,"capability_non_primary_group_count":37},{"model":"GPT-4 Preview (1106)","provider":"OpenAI","country":"US","openness":"api","release_date":"2023-11-06","param_count":1760.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":61,"benchmark_group_count":47,"benchmark_coverage_score":32.0,"source_family_count":18,"source_group_family_count":18,"source_effective_breadth":6.8,"source_primary":"euroeval.com","source_primary_share":0.3607,"source_primary_group_share":0.2766,"capability_breadth":4.94,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3607,"capability_primary_group_count":15,"capability_non_primary_group_count":32},{"model":"Qwen 3 32B (Thinking)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":32.0,"active_param_count":null,"training_compute_flop":7.085e+24,"num_benchmarks":61,"benchmark_group_count":33,"benchmark_coverage_score":17.95,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.61,"source_primary":"artificialanalysis.ai","source_primary_share":0.623,"source_primary_group_share":0.5758,"capability_breadth":4.49,"capability_primary":"knowledge_general","capability_primary_share":0.3279,"capability_primary_group_count":9,"capability_non_primary_group_count":24},{"model":"MiMo-V2-Flash (Non-reasoning)","provider":"Xiaomi","country":"CN","openness":"open","release_date":"2026-01-01","param_count":309.0,"active_param_count":15.0,"training_compute_flop":2.43e+24,"num_benchmarks":61,"benchmark_group_count":27,"benchmark_coverage_score":17.69,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.6,"source_primary":"artificialanalysis.ai","source_primary_share":0.9,"source_primary_group_share":0.7778,"capability_breadth":4.42,"capability_primary":"knowledge_general","capability_primary_share":0.3,"capability_primary_group_count":7,"capability_non_primary_group_count":20},{"model":"Mixtral 8x22B Instruct (v0.1)","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-04-17","param_count":141.0,"active_param_count":39.0,"training_compute_flop":2.34e+24,"num_benchmarks":60,"benchmark_group_count":53,"benchmark_coverage_score":42.0,"source_family_count":24,"source_group_family_count":24,"source_effective_breadth":7.3,"source_primary":"livebench.ai","source_primary_share":0.2881,"source_primary_group_share":0.3208,"capability_breadth":7.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2034,"capability_primary_group_count":11,"capability_non_primary_group_count":42},{"model":"GPT-3.5 Turbo (0613)","provider":"OpenAI","country":"US","openness":"api","release_date":"2023-06-13","param_count":175.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI GPT-3/GPT-3.5 family retired or superseded by newer API models.","num_benchmarks":60,"benchmark_group_count":54,"benchmark_coverage_score":36.0,"source_family_count":15,"source_group_family_count":15,"source_effective_breadth":3.54,"source_primary":"crfm.stanford.edu","source_primary_share":0.4655,"source_primary_group_share":0.5,"capability_breadth":6.12,"capability_primary":"knowledge_general","capability_primary_share":0.3103,"capability_primary_group_count":18,"capability_non_primary_group_count":36},{"model":"Yi 34B (Chat)","provider":"01.AI","country":"CN","openness":"open","release_date":"2023-11-02","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":60,"benchmark_group_count":53,"benchmark_coverage_score":40.0,"source_family_count":19,"source_group_family_count":19,"source_effective_breadth":10.68,"source_primary":"huggingface.co/spaces/nphardeval/nphardeval-leaderboard","source_primary_share":0.1724,"source_primary_group_share":0.1509,"capability_breadth":5.72,"capability_primary":"reasoning_logical","capability_primary_share":0.2586,"capability_primary_group_count":13,"capability_non_primary_group_count":40},{"model":"GPT-5.5 (Medium)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-04-23","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":60,"benchmark_group_count":39,"benchmark_coverage_score":19.56,"source_family_count":17,"source_group_family_count":17,"source_effective_breadth":5.05,"source_primary":"artificialanalysis.ai","source_primary_share":0.6102,"source_primary_group_share":0.4103,"capability_breadth":4.89,"capability_primary":"code","capability_primary_share":0.3276,"capability_primary_group_count":10,"capability_non_primary_group_count":28},{"model":"MiMo-V2-Flash (Reasoning)","provider":"Xiaomi","country":"CN","openness":"open","release_date":"2026-01-01","param_count":309.0,"active_param_count":15.0,"training_compute_flop":2.43e+24,"num_benchmarks":60,"benchmark_group_count":27,"benchmark_coverage_score":17.43,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.76,"source_primary":"artificialanalysis.ai","source_primary_share":0.8644,"source_primary_group_share":0.7407,"capability_breadth":4.36,"capability_primary":"knowledge_general","capability_primary_share":0.3051,"capability_primary_group_count":7,"capability_non_primary_group_count":20},{"model":"Nova Premier","provider":"Amazon","country":"US","openness":"api","release_date":"2024-12-03","param_count":2000.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":60,"benchmark_group_count":29,"benchmark_coverage_score":19.1,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":1.82,"source_primary":"artificialanalysis.ai","source_primary_share":0.8667,"source_primary_group_share":0.7241,"capability_breadth":4.77,"capability_primary":"code","capability_primary_share":0.2833,"capability_primary_group_count":8,"capability_non_primary_group_count":21},{"model":"KAT-Coder-Pro V1","provider":"Kuaishou","country":"CN","openness":"api","release_date":"2025-10-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":60,"benchmark_group_count":27,"benchmark_coverage_score":17.78,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.61,"source_primary":"artificialanalysis.ai","source_primary_share":0.9,"source_primary_group_share":0.7778,"capability_breadth":4.45,"capability_primary":"code","capability_primary_share":0.3051,"capability_primary_group_count":9,"capability_non_primary_group_count":17},{"model":"Claude Opus 4.1 (20250805)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-08-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":60,"benchmark_group_count":53,"benchmark_coverage_score":39.0,"source_family_count":16,"source_group_family_count":16,"source_effective_breadth":6.49,"source_primary":"vals.ai","source_primary_share":0.2586,"source_primary_group_share":0.283,"capability_breadth":5.76,"capability_primary":"knowledge_general","capability_primary_share":0.2586,"capability_primary_group_count":14,"capability_non_primary_group_count":39},{"model":"GPT-5.2 Pro","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-12-11","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":59,"benchmark_group_count":49,"benchmark_coverage_score":32.0,"source_family_count":15,"source_group_family_count":15,"source_effective_breadth":5.87,"source_primary":"ppbench.com","source_primary_share":0.3036,"source_primary_group_share":0.3265,"capability_breadth":5.06,"capability_primary":"reasoning_logical","capability_primary_share":0.3214,"capability_primary_group_count":17,"capability_non_primary_group_count":32},{"model":"Claude 2","provider":"Anthropic","country":"US","openness":"api","release_date":"2023-07-11","param_count":null,"active_param_count":null,"training_compute_flop":3.866e+24,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 1/2 family retired or superseded by newer Claude models.","num_benchmarks":59,"benchmark_group_count":50,"benchmark_coverage_score":36.0,"source_family_count":18,"source_group_family_count":18,"source_effective_breadth":10.5,"source_primary":"huggingface.co/spaces/nphardeval/nphardeval-leaderboard","source_primary_share":0.1724,"source_primary_group_share":0.16,"capability_breadth":6.05,"capability_primary":"knowledge_general","capability_primary_share":0.2586,"capability_primary_group_count":14,"capability_non_primary_group_count":36},{"model":"Yi 1.5 34B Chat","provider":"01.AI","country":"CN","openness":"open","release_date":"2024-05-13","param_count":34.0,"active_param_count":null,"training_compute_flop":7.344e+23,"num_benchmarks":59,"benchmark_group_count":53,"benchmark_coverage_score":36.0,"source_family_count":23,"source_group_family_count":23,"source_effective_breadth":10.52,"source_primary":"huggingface.co/spaces/xufengduan/humanlikeness","source_primary_share":0.1724,"source_primary_group_share":0.1887,"capability_breadth":4.53,"capability_primary":"reasoning_logical","capability_primary_share":0.3103,"capability_primary_group_count":17,"capability_non_primary_group_count":36},{"model":"GPT-5.1 Codex (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-11-19","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":59,"benchmark_group_count":27,"benchmark_coverage_score":16.76,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.72,"source_primary":"artificialanalysis.ai","source_primary_share":0.8814,"source_primary_group_share":0.7407,"capability_breadth":4.19,"capability_primary":"knowledge_general","capability_primary_share":0.3051,"capability_primary_group_count":8,"capability_non_primary_group_count":19},{"model":"Claude Sonnet 4.6 (Adaptive Reasoning, Max Effort)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-02-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":59,"benchmark_group_count":25,"benchmark_coverage_score":18.7,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.39,"source_primary":"artificialanalysis.ai","source_primary_share":0.931,"source_primary_group_share":0.84,"capability_breadth":4.67,"capability_primary":"knowledge_general","capability_primary_share":0.2807,"capability_primary_group_count":5,"capability_non_primary_group_count":19},{"model":"Grok 4.20 0309 (Non-reasoning)","provider":"xAI","country":"US","openness":"api","release_date":"2026-03-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":59,"benchmark_group_count":27,"benchmark_coverage_score":19.79,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":2.15,"source_primary":"artificialanalysis.ai","source_primary_share":0.8448,"source_primary_group_share":0.6667,"capability_breadth":4.95,"capability_primary":"knowledge_general","capability_primary_share":0.2759,"capability_primary_group_count":5,"capability_non_primary_group_count":22},{"model":"Claude Opus 4.5 (Non-reasoning)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-11-24","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":58,"benchmark_group_count":26,"benchmark_coverage_score":19.67,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.64,"source_primary":"artificialanalysis.ai","source_primary_share":0.8966,"source_primary_group_share":0.7692,"capability_breadth":4.92,"capability_primary":"knowledge_general","capability_primary_share":0.2586,"capability_primary_group_count":6,"capability_non_primary_group_count":20},{"model":"Llama 2 70B","provider":"Meta","country":"US","openness":"open","release_date":"2023-07-18","param_count":70.0,"active_param_count":null,"training_compute_flop":8.1e+23,"num_benchmarks":58,"benchmark_group_count":52,"benchmark_coverage_score":33.0,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":3.28,"source_primary":"crfm.stanford.edu","source_primary_share":0.4655,"source_primary_group_share":0.5192,"capability_breadth":5.11,"capability_primary":"knowledge_general","capability_primary_share":0.3448,"capability_primary_group_count":19,"capability_non_primary_group_count":33},{"model":"GPT-5.6 Sol (Max)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":58,"benchmark_group_count":38,"benchmark_coverage_score":19.96,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":4.01,"source_primary":"artificialanalysis.ai","source_primary_share":0.6207,"source_primary_group_share":0.4211,"capability_breadth":4.99,"capability_primary":"code","capability_primary_share":0.3333,"capability_primary_group_count":10,"capability_non_primary_group_count":27},{"model":"DeepSeek V3.1 Terminus (Reasoning)","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-09-22","param_count":671.0,"active_param_count":37.0,"training_compute_flop":3.594058e+24,"num_benchmarks":58,"benchmark_group_count":24,"benchmark_coverage_score":17.4,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"artificialanalysis.ai","source_primary_share":0.9474,"source_primary_group_share":0.875,"capability_breadth":4.35,"capability_primary":"knowledge_general","capability_primary_share":0.2982,"capability_primary_group_count":6,"capability_non_primary_group_count":18},{"model":"GLM-4.6 (Reasoning)","provider":"Zhipu","country":"CN","openness":"open","release_date":"2025-09-30","param_count":357.0,"active_param_count":32.0,"training_compute_flop":4.42e+24,"num_benchmarks":58,"benchmark_group_count":24,"benchmark_coverage_score":17.4,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"artificialanalysis.ai","source_primary_share":0.9474,"source_primary_group_share":0.875,"capability_breadth":4.35,"capability_primary":"knowledge_general","capability_primary_share":0.2982,"capability_primary_group_count":6,"capability_non_primary_group_count":18},{"model":"GPT-5 Chat","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-08-07","param_count":null,"active_param_count":null,"training_compute_flop":6.6e+25,"num_benchmarks":58,"benchmark_group_count":52,"benchmark_coverage_score":42.0,"source_family_count":23,"source_group_family_count":23,"source_effective_breadth":10.73,"source_primary":"artificialanalysis.ai","source_primary_share":0.193,"source_primary_group_share":0.1923,"capability_breadth":7.91,"capability_primary":"reasoning_logical","capability_primary_share":0.193,"capability_primary_group_count":10,"capability_non_primary_group_count":42},{"model":"Hermes-4-14B","provider":"Nous Research","country":"US","openness":"open","release_date":"2025-08-26","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":58,"benchmark_group_count":37,"benchmark_coverage_score":12.62,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.21,"source_primary":"medmarks.ai","source_primary_share":0.4655,"source_primary_group_share":0.5676,"capability_breadth":3.16,"capability_primary":"knowledge_general","capability_primary_share":0.4655,"capability_primary_group_count":20,"capability_non_primary_group_count":17},{"model":"Median Human","provider":"Human","country":null,"openness":null,"release_date":null,"param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":57,"benchmark_group_count":50,"benchmark_coverage_score":35.0,"source_family_count":23,"source_group_family_count":23,"source_effective_breadth":7.81,"source_primary":"epoch.ai","source_primary_share":0.2642,"source_primary_group_share":0.28,"capability_breadth":4.44,"capability_primary":"vision_language","capability_primary_share":0.3019,"capability_primary_group_count":15,"capability_non_primary_group_count":35},{"model":"Command-R+ (08-2024)","provider":"Cohere","country":"CA","openness":"open","release_date":"2024-08-21","param_count":104.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":57,"benchmark_group_count":50,"benchmark_coverage_score":38.0,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":5.25,"source_primary":"livebench.ai","source_primary_share":0.3091,"source_primary_group_share":0.34,"capability_breadth":6.01,"capability_primary":"reasoning_logical","capability_primary_share":0.2364,"capability_primary_group_count":12,"capability_non_primary_group_count":38},{"model":"Claude Mythos 5","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-06-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":57,"benchmark_group_count":57,"benchmark_coverage_score":41.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.67,"source_primary":"benchmarklist.com","source_primary_share":0.4386,"source_primary_group_share":0.4386,"capability_breadth":6.03,"capability_primary":"knowledge_general","capability_primary_share":0.2807,"capability_primary_group_count":16,"capability_non_primary_group_count":41},{"model":"GPT-5.5 Instant","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-04-23","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":57,"benchmark_group_count":34,"benchmark_coverage_score":18.83,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":4.22,"source_primary":"artificialanalysis.ai","source_primary_share":0.625,"source_primary_group_share":0.4412,"capability_breadth":4.71,"capability_primary":"knowledge_general","capability_primary_share":0.3214,"capability_primary_group_count":10,"capability_non_primary_group_count":24},{"model":"Mercury 2","provider":"Inception","country":"UAE","openness":"api","release_date":"2026-02-24","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":57,"benchmark_group_count":35,"benchmark_coverage_score":16.42,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":4.45,"source_primary":"artificialanalysis.ai","source_primary_share":0.6071,"source_primary_group_share":0.4286,"capability_breadth":4.1,"capability_primary":"code","capability_primary_share":0.4107,"capability_primary_group_count":13,"capability_non_primary_group_count":22},{"model":"Doubao Seed Code","provider":"ByteDance","country":"CN","openness":"api","release_date":"2025-12-01","param_count":1000.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":57,"benchmark_group_count":24,"benchmark_coverage_score":18.18,"source_family_count":4,"source_group_family_count":3,"source_effective_breadth":1.4,"source_primary":"artificialanalysis.ai","source_primary_share":0.9123,"source_primary_group_share":0.8333,"capability_breadth":4.54,"capability_primary":"code","capability_primary_share":0.2807,"capability_primary_group_count":7,"capability_non_primary_group_count":17},{"model":"GLM-4.7 (Non-reasoning)","provider":"Zhipu","country":"CN","openness":"open","release_date":"2025-12-22","param_count":355.0,"active_param_count":32.0,"training_compute_flop":4.42e+24,"num_benchmarks":57,"benchmark_group_count":25,"benchmark_coverage_score":16.55,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.52,"source_primary":"artificialanalysis.ai","source_primary_share":0.9107,"source_primary_group_share":0.8,"capability_breadth":4.14,"capability_primary":"knowledge_general","capability_primary_share":0.3036,"capability_primary_group_count":7,"capability_non_primary_group_count":18},{"model":"falcon-40B","provider":"TII","country":"UAE","openness":"open","release_date":"2023-05-25","param_count":40.0,"active_param_count":null,"training_compute_flop":2.4e+23,"num_benchmarks":56,"benchmark_group_count":53,"benchmark_coverage_score":14.79,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":3.11,"source_primary":"crfm.stanford.edu","source_primary_share":0.4821,"source_primary_group_share":0.5094,"capability_breadth":3.7,"capability_primary":"knowledge_general","capability_primary_share":0.4643,"capability_primary_group_count":25,"capability_non_primary_group_count":28},{"model":"Qwen 3.5 122B A10B (Non-reasoning)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-02-24","param_count":122.0,"active_param_count":10.0,"training_compute_flop":null,"num_benchmarks":56,"benchmark_group_count":22,"benchmark_coverage_score":17.77,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.31,"source_primary":"artificialanalysis.ai","source_primary_share":0.9455,"source_primary_group_share":0.8636,"capability_breadth":4.44,"capability_primary":"knowledge_general","capability_primary_share":0.2909,"capability_primary_group_count":5,"capability_non_primary_group_count":17},{"model":"Qwen 3.5 122B A10B (Reasoning)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-02-24","param_count":122.0,"active_param_count":10.0,"training_compute_flop":null,"num_benchmarks":56,"benchmark_group_count":22,"benchmark_coverage_score":17.77,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.31,"source_primary":"artificialanalysis.ai","source_primary_share":0.9455,"source_primary_group_share":0.8636,"capability_breadth":4.44,"capability_primary":"knowledge_general","capability_primary_share":0.2909,"capability_primary_group_count":5,"capability_non_primary_group_count":17},{"model":"Qwen 3.5 35B A3B (Non-reasoning)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-02-24","param_count":35.0,"active_param_count":3.0,"training_compute_flop":null,"num_benchmarks":56,"benchmark_group_count":22,"benchmark_coverage_score":17.77,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.31,"source_primary":"artificialanalysis.ai","source_primary_share":0.9455,"source_primary_group_share":0.8636,"capability_breadth":4.44,"capability_primary":"knowledge_general","capability_primary_share":0.2909,"capability_primary_group_count":5,"capability_non_primary_group_count":17},{"model":"GLM-4.6V (Reasoning)","provider":"Zhipu","country":"CN","openness":"open","release_date":"2025-12-09","param_count":106.0,"active_param_count":12.0,"training_compute_flop":null,"num_benchmarks":56,"benchmark_group_count":23,"benchmark_coverage_score":16.64,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.29,"source_primary":"artificialanalysis.ai","source_primary_share":0.9455,"source_primary_group_share":0.8696,"capability_breadth":4.16,"capability_primary":"knowledge_general","capability_primary_share":0.3091,"capability_primary_group_count":6,"capability_non_primary_group_count":17},{"model":"GLM-4.6V (Non-reasoning)","provider":"Zhipu","country":"CN","openness":"open","release_date":"2025-12-09","param_count":106.0,"active_param_count":12.0,"training_compute_flop":null,"num_benchmarks":56,"benchmark_group_count":23,"benchmark_coverage_score":16.64,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.29,"source_primary":"artificialanalysis.ai","source_primary_share":0.9455,"source_primary_group_share":0.8696,"capability_breadth":4.16,"capability_primary":"knowledge_general","capability_primary_share":0.3091,"capability_primary_group_count":6,"capability_non_primary_group_count":17},{"model":"Claude Opus 4.5 (20251101)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-11-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":55,"benchmark_group_count":50,"benchmark_coverage_score":37.0,"source_family_count":16,"source_group_family_count":16,"source_effective_breadth":5.76,"source_primary":"vals.ai","source_primary_share":0.3208,"source_primary_group_share":0.34,"capability_breadth":6.87,"capability_primary":"knowledge_general","capability_primary_share":0.2453,"capability_primary_group_count":13,"capability_non_primary_group_count":37},{"model":"granite-3.0-8B-instruct","provider":"IBM","country":"US","openness":"open","release_date":"2024-10-21","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":54,"benchmark_group_count":34,"benchmark_coverage_score":8.99,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.03,"source_primary":"euroeval.com","source_primary_share":0.7407,"source_primary_group_share":0.6765,"capability_breadth":2.25,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6296,"capability_primary_group_count":22,"capability_non_primary_group_count":12},{"model":"text-davinci-003","provider":"OpenAI","country":"US","openness":"api","release_date":"2022-11-28","param_count":175.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI completions family retired or superseded by newer API models.","num_benchmarks":54,"benchmark_group_count":46,"benchmark_coverage_score":32.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.77,"source_primary":"crfm.stanford.edu","source_primary_share":0.566,"source_primary_group_share":0.5652,"capability_breadth":4.83,"capability_primary":"knowledge_general","capability_primary_share":0.3208,"capability_primary_group_count":14,"capability_non_primary_group_count":32},{"model":"MiMo-V2-Omni","provider":"Xiaomi","country":"CN","openness":"open","release_date":"2026-03-18","param_count":309.0,"active_param_count":15.0,"training_compute_flop":null,"num_benchmarks":54,"benchmark_group_count":34,"benchmark_coverage_score":20.61,"source_family_count":14,"source_group_family_count":13,"source_effective_breadth":4.55,"source_primary":"artificialanalysis.ai","source_primary_share":0.5926,"source_primary_group_share":0.4118,"capability_breadth":5.15,"capability_primary":"code","capability_primary_share":0.2963,"capability_primary_group_count":7,"capability_non_primary_group_count":27},{"model":"JT-MINI","provider":"China Mobile","country":"CN","openness":"api","release_date":"2026-04-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":54,"benchmark_group_count":30,"benchmark_coverage_score":17.78,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.27,"source_primary":"artificialanalysis.ai","source_primary_share":0.5741,"source_primary_group_share":0.4667,"capability_breadth":4.45,"capability_primary":"knowledge_general","capability_primary_share":0.3333,"capability_primary_group_count":9,"capability_non_primary_group_count":21},{"model":"Qwen 3 Coder 480B A35B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-07-23","param_count":480.0,"active_param_count":35.0,"training_compute_flop":1.575e+24,"num_benchmarks":54,"benchmark_group_count":35,"benchmark_coverage_score":12.87,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":3.45,"source_primary":"artificialanalysis.ai","source_primary_share":0.6481,"source_primary_group_share":0.5143,"capability_breadth":3.22,"capability_primary":"code","capability_primary_share":0.463,"capability_primary_group_count":15,"capability_non_primary_group_count":20},{"model":"gpt-sw3-20B-instruct","provider":"AI Sweden","country":"SE","openness":"open","release_date":"2023-04-28","param_count":20.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":54,"benchmark_group_count":31,"benchmark_coverage_score":7.58,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.37,"source_primary":"euroeval.com","source_primary_share":0.8704,"source_primary_group_share":0.8387,"capability_breadth":1.9,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7037,"capability_primary_group_count":23,"capability_non_primary_group_count":8},{"model":"Gemini 1.5 Flash (001)","provider":"Google","country":"US","openness":"api","release_date":"2024-05-15","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 1.x family retired or superseded by newer Gemini models.","num_benchmarks":53,"benchmark_group_count":47,"benchmark_coverage_score":34.0,"source_family_count":14,"source_group_family_count":14,"source_effective_breadth":7.64,"source_primary":"crfm.stanford.edu","source_primary_share":0.2642,"source_primary_group_share":0.234,"capability_breadth":4.97,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3208,"capability_primary_group_count":13,"capability_non_primary_group_count":34},{"model":"Qwen 3.5 397B A17B (Non-reasoning)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-02-16","param_count":397.0,"active_param_count":17.0,"training_compute_flop":null,"num_benchmarks":53,"benchmark_group_count":21,"benchmark_coverage_score":16.39,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"artificialanalysis.ai","source_primary_share":0.9423,"source_primary_group_share":0.8571,"capability_breadth":4.1,"capability_primary":"knowledge_general","capability_primary_share":0.3077,"capability_primary_group_count":5,"capability_non_primary_group_count":16},{"model":"Qwen 3 Max (Preview)","provider":"Alibaba","country":"CN","openness":"api","release_date":"2025-09-05","param_count":700.0,"active_param_count":null,"training_compute_flop":1.512e+25,"num_benchmarks":53,"benchmark_group_count":36,"benchmark_coverage_score":15.29,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":3.09,"source_primary":"artificialanalysis.ai","source_primary_share":0.6415,"source_primary_group_share":0.4722,"capability_breadth":3.82,"capability_primary":"knowledge_general","capability_primary_share":0.3585,"capability_primary_group_count":11,"capability_non_primary_group_count":25},{"model":"Qwen 3.5 27B (Non-reasoning)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-02-24","param_count":27.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":53,"benchmark_group_count":21,"benchmark_coverage_score":16.39,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"artificialanalysis.ai","source_primary_share":0.9423,"source_primary_group_share":0.8571,"capability_breadth":4.1,"capability_primary":"knowledge_general","capability_primary_share":0.3077,"capability_primary_group_count":5,"capability_non_primary_group_count":16},{"model":"Command R","provider":"Cohere","country":"CA","openness":"open","release_date":"2024-03-11","param_count":35.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":52,"benchmark_group_count":46,"benchmark_coverage_score":36.0,"source_family_count":17,"source_group_family_count":17,"source_effective_breadth":5.21,"source_primary":"livebench.ai","source_primary_share":0.3269,"source_primary_group_share":0.3696,"capability_breadth":6.93,"capability_primary":"knowledge_general","capability_primary_share":0.2115,"capability_primary_group_count":10,"capability_non_primary_group_count":36},{"model":"LFM2 8B A1B","provider":"Liquid AI","country":"US","openness":"open","release_date":"2025-11-28","param_count":8.0,"active_param_count":1.0,"training_compute_flop":null,"num_benchmarks":52,"benchmark_group_count":27,"benchmark_coverage_score":14.76,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.33,"source_primary":"artificialanalysis.ai","source_primary_share":0.6667,"source_primary_group_share":0.6296,"capability_breadth":3.69,"capability_primary":"knowledge_general","capability_primary_share":0.4118,"capability_primary_group_count":7,"capability_non_primary_group_count":20},{"model":"K-EXAONE (Non-reasoning)","provider":"LG AI","country":"KR","openness":"open","release_date":"2026-01-12","param_count":236.0,"active_param_count":23.0,"training_compute_flop":1.52e+24,"num_benchmarks":52,"benchmark_group_count":21,"benchmark_coverage_score":16.85,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.21,"source_primary":"artificialanalysis.ai","source_primary_share":0.9615,"source_primary_group_share":0.9048,"capability_breadth":4.21,"capability_primary":"knowledge_general","capability_primary_share":0.2885,"capability_primary_group_count":5,"capability_non_primary_group_count":16},{"model":"Qwen 1.5 72B Chat","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-02-05","param_count":72.0,"active_param_count":null,"training_compute_flop":1.3e+24,"num_benchmarks":51,"benchmark_group_count":47,"benchmark_coverage_score":38.0,"source_family_count":18,"source_group_family_count":18,"source_effective_breadth":5.56,"source_primary":"livebench.ai","source_primary_share":0.3333,"source_primary_group_share":0.3617,"capability_breadth":7.33,"capability_primary":"knowledge_multilingual","capability_primary_share":0.1765,"capability_primary_group_count":9,"capability_non_primary_group_count":38},{"model":"Yi 34B (Base)","provider":"01.AI","country":"CN","openness":"open","release_date":"2023-11-02","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":51,"benchmark_group_count":38,"benchmark_coverage_score":20.13,"source_family_count":13,"source_group_family_count":12,"source_effective_breadth":8.3,"source_primary":"huggingface.co/spaces/allenai/urial-bench","source_primary_share":0.2245,"source_primary_group_share":0.1842,"capability_breadth":5.03,"capability_primary":"knowledge_general","capability_primary_share":0.3469,"capability_primary_group_count":14,"capability_non_primary_group_count":24},{"model":"c4ai-command-r7B-12-2024","provider":"Cohere","country":"CA","openness":"open","release_date":"2024-12-11","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":51,"benchmark_group_count":33,"benchmark_coverage_score":13.82,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":3.41,"source_primary":"euroeval.com","source_primary_share":0.4902,"source_primary_group_share":0.4242,"capability_breadth":3.45,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4706,"capability_primary_group_count":15,"capability_non_primary_group_count":18},{"model":"GPT-5.6 Terra (Max)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":51,"benchmark_group_count":31,"benchmark_coverage_score":18.28,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.87,"source_primary":"artificialanalysis.ai","source_primary_share":0.7059,"source_primary_group_share":0.5161,"capability_breadth":4.57,"capability_primary":"code","capability_primary_share":0.3529,"capability_primary_group_count":9,"capability_non_primary_group_count":22},{"model":"Grok 4.5 (High)","provider":"xAI","country":"US","openness":"api","release_date":"2026-07-08","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":51,"benchmark_group_count":32,"benchmark_coverage_score":16.34,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":3.94,"source_primary":"artificialanalysis.ai","source_primary_share":0.6471,"source_primary_group_share":0.4375,"capability_breadth":4.08,"capability_primary":"code","capability_primary_share":0.4,"capability_primary_group_count":11,"capability_non_primary_group_count":20},{"model":"Grok 4.3 (High)","provider":"xAI","country":"US","openness":"api","release_date":"2026-04-30","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":51,"benchmark_group_count":30,"benchmark_coverage_score":18.87,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.66,"source_primary":"artificialanalysis.ai","source_primary_share":0.7255,"source_primary_group_share":0.5667,"capability_breadth":4.72,"capability_primary":"code","capability_primary_share":0.34,"capability_primary_group_count":8,"capability_non_primary_group_count":21},{"model":"GPT-5.4 (Low)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-03-06","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":51,"benchmark_group_count":30,"benchmark_coverage_score":20.24,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":3.88,"source_primary":"artificialanalysis.ai","source_primary_share":0.64,"source_primary_group_share":0.4667,"capability_breadth":5.06,"capability_primary":"code","capability_primary_share":0.32,"capability_primary_group_count":6,"capability_non_primary_group_count":24},{"model":"Nova 2.0 Pro Preview (Low)","provider":"Amazon","country":"US","openness":"api","release_date":"2025-08-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":51,"benchmark_group_count":20,"benchmark_coverage_score":18.55,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.22,"source_primary":"artificialanalysis.ai","source_primary_share":0.9608,"source_primary_group_share":0.9,"capability_breadth":4.64,"capability_primary":"knowledge_general","capability_primary_share":0.2941,"capability_primary_group_count":6,"capability_non_primary_group_count":14},{"model":"Qwen 3 Coder 30B A3B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-07-23","param_count":30.0,"active_param_count":3.0,"training_compute_flop":null,"num_benchmarks":51,"benchmark_group_count":32,"benchmark_coverage_score":15.29,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.78,"source_primary":"artificialanalysis.ai","source_primary_share":0.7,"source_primary_group_share":0.5625,"capability_breadth":3.82,"capability_primary":"code","capability_primary_share":0.4,"capability_primary_group_count":10,"capability_non_primary_group_count":22},{"model":"Mistral 7B","provider":"Mistral","country":"FR","openness":"open","release_date":"2023-09-27","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":50,"benchmark_group_count":44,"benchmark_coverage_score":30.0,"source_family_count":15,"source_group_family_count":14,"source_effective_breadth":5.76,"source_primary":"benchmarklist.com","source_primary_share":0.26,"source_primary_group_share":0.2955,"capability_breadth":5.43,"capability_primary":"knowledge_general","capability_primary_share":0.28,"capability_primary_group_count":14,"capability_non_primary_group_count":30},{"model":"GPT-5.1 Instant","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-11-12","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":50,"benchmark_group_count":31,"benchmark_coverage_score":18.09,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":2.77,"source_primary":"artificialanalysis.ai","source_primary_share":0.7143,"source_primary_group_share":0.5484,"capability_breadth":4.52,"capability_primary":"code","capability_primary_share":0.3265,"capability_primary_group_count":7,"capability_non_primary_group_count":24},{"model":"Qwen2.5-Omni-7B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-03-26","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":50,"benchmark_group_count":45,"benchmark_coverage_score":7.06,"source_family_count":9,"source_group_family_count":8,"source_effective_breadth":4.72,"source_primary":"llm-stats.com","source_primary_share":0.38,"source_primary_group_share":0.3333,"capability_breadth":1.77,"capability_primary":"vision_language","capability_primary_share":0.74,"capability_primary_group_count":34,"capability_non_primary_group_count":11},{"model":"Phi-3.5-MoE-instruct","provider":"Microsoft","country":"US","openness":"open","release_date":"2024-08-23","param_count":42.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":50,"benchmark_group_count":40,"benchmark_coverage_score":26.04,"source_family_count":11,"source_group_family_count":10,"source_effective_breadth":4.26,"source_primary":"livebench.ai","source_primary_share":0.34,"source_primary_group_share":0.425,"capability_breadth":6.51,"capability_primary":"reasoning_logical","capability_primary_share":0.24,"capability_primary_group_count":11,"capability_non_primary_group_count":29},{"model":"Gemini 2.5 Flash Lite (Preview 09-2025)","provider":"Google","country":"US","openness":"api","release_date":"2025-09-25","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 2.5 preview snapshot retired or superseded by newer Gemini models.","num_benchmarks":50,"benchmark_group_count":33,"benchmark_coverage_score":16.88,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":3.96,"source_primary":"huggingface.co/spaces/mlx-community/mlx-benchmark-leaderboard","source_primary_share":0.3878,"source_primary_group_share":0.4242,"capability_breadth":4.22,"capability_primary":"code","capability_primary_share":0.3265,"capability_primary_group_count":12,"capability_non_primary_group_count":21},{"model":"Grok Build 0.1","provider":"xAI","country":"US","openness":"api","release_date":"2026-05-29","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":50,"benchmark_group_count":30,"benchmark_coverage_score":14.02,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":4.29,"source_primary":"artificialanalysis.ai","source_primary_share":0.6531,"source_primary_group_share":0.4333,"capability_breadth":3.51,"capability_primary":"code","capability_primary_share":0.449,"capability_primary_group_count":13,"capability_non_primary_group_count":17},{"model":"GPT-5.5 (Low)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-04-23","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":50,"benchmark_group_count":29,"benchmark_coverage_score":19.56,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":3.22,"source_primary":"artificialanalysis.ai","source_primary_share":0.7143,"source_primary_group_share":0.5172,"capability_breadth":4.89,"capability_primary":"code","capability_primary_share":0.3265,"capability_primary_group_count":7,"capability_non_primary_group_count":22},{"model":"Ling-2.6-flash","provider":"InclusionAI","country":"CN","openness":"open","release_date":"2026-04-22","param_count":104.0,"active_param_count":7.4,"training_compute_flop":null,"num_benchmarks":50,"benchmark_group_count":29,"benchmark_coverage_score":19.53,"source_family_count":10,"source_group_family_count":9,"source_effective_breadth":3.25,"source_primary":"artificialanalysis.ai","source_primary_share":0.68,"source_primary_group_share":0.5172,"capability_breadth":4.88,"capability_primary":"code","capability_primary_share":0.34,"capability_primary_group_count":8,"capability_non_primary_group_count":21},{"model":"granite-3.0-1B-a400m-instruct","provider":"IBM","country":"US","openness":"open","release_date":"2024-10-03","param_count":1.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":50,"benchmark_group_count":29,"benchmark_coverage_score":8.99,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.49,"source_primary":"euroeval.com","source_primary_share":0.88,"source_primary_group_share":0.7931,"capability_breadth":2.25,"capability_primary":"knowledge_multilingual","capability_primary_share":0.62,"capability_primary_group_count":20,"capability_non_primary_group_count":9},{"model":"granite-3.0-1B-a400m-base","provider":"IBM","country":"US","openness":"open","release_date":"2024-10-03","param_count":1.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":50,"benchmark_group_count":29,"benchmark_coverage_score":8.99,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.49,"source_primary":"euroeval.com","source_primary_share":0.88,"source_primary_group_share":0.7931,"capability_breadth":2.25,"capability_primary":"knowledge_multilingual","capability_primary_share":0.62,"capability_primary_group_count":20,"capability_non_primary_group_count":9},{"model":"Claude Opus 4.8 (Max)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-05-29","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":49,"benchmark_group_count":42,"benchmark_coverage_score":10.87,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":4.14,"source_primary":"matharena.ai","source_primary_share":0.4167,"source_primary_group_share":0.3571,"capability_breadth":2.72,"capability_primary":"math","capability_primary_share":0.5833,"capability_primary_group_count":23,"capability_non_primary_group_count":19},{"model":"Qwen 3.5 Plus","provider":"Alibaba","country":"CN","openness":"api","release_date":"2026-02-16","param_count":397.0,"active_param_count":17.0,"training_compute_flop":null,"num_benchmarks":49,"benchmark_group_count":33,"benchmark_coverage_score":22.92,"source_family_count":14,"source_group_family_count":13,"source_effective_breadth":6.6,"source_primary":"fab-ai.org","source_primary_share":0.4286,"source_primary_group_share":0.2424,"capability_breadth":5.73,"capability_primary":"vision_language","capability_primary_share":0.3265,"capability_primary_group_count":10,"capability_non_primary_group_count":23},{"model":"aya-vision-8B","provider":"Cohere","country":"CA","openness":"open","release_date":"2025-03-02","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":49,"benchmark_group_count":40,"benchmark_coverage_score":12.55,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.0,"source_primary":"huggingface.co","source_primary_share":0.6122,"source_primary_group_share":0.65,"capability_breadth":3.14,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3878,"capability_primary_group_count":16,"capability_non_primary_group_count":24},{"model":"zephyr-7B-alpha","provider":"HuggingFace","country":"FR","openness":"open","release_date":"2023-10-09","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":49,"benchmark_group_count":45,"benchmark_coverage_score":19.53,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":4.21,"source_primary":"livebench.ai","source_primary_share":0.3542,"source_primary_group_share":0.3778,"capability_breadth":4.88,"capability_primary":"reasoning_logical","capability_primary_share":0.375,"capability_primary_group_count":18,"capability_non_primary_group_count":27},{"model":"mistral-nemo-minitron-8B-base","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-08-19","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":49,"benchmark_group_count":27,"benchmark_coverage_score":9.14,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.82,"source_primary":"euroeval.com","source_primary_share":0.7959,"source_primary_group_share":0.7037,"capability_breadth":2.28,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6122,"capability_primary_group_count":18,"capability_non_primary_group_count":9},{"model":"GPT-5.6 Luna (Max)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":49,"benchmark_group_count":30,"benchmark_coverage_score":17.24,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.26,"source_primary":"artificialanalysis.ai","source_primary_share":0.6735,"source_primary_group_share":0.4667,"capability_breadth":4.31,"capability_primary":"code","capability_primary_share":0.3673,"capability_primary_group_count":9,"capability_non_primary_group_count":21},{"model":"Gemini 3.5 Flash (Medium)","provider":"Google","country":"US","openness":"api","release_date":"2026-05-19","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":49,"benchmark_group_count":29,"benchmark_coverage_score":17.57,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.83,"source_primary":"artificialanalysis.ai","source_primary_share":0.6875,"source_primary_group_share":0.5172,"capability_breadth":4.39,"capability_primary":"code","capability_primary_share":0.3404,"capability_primary_group_count":9,"capability_non_primary_group_count":19},{"model":"GPT-5.4 Nano (xHigh)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-03-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":49,"benchmark_group_count":29,"benchmark_coverage_score":16.47,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":2.93,"source_primary":"artificialanalysis.ai","source_primary_share":0.7347,"source_primary_group_share":0.5517,"capability_breadth":4.12,"capability_primary":"code","capability_primary_share":0.3878,"capability_primary_group_count":10,"capability_non_primary_group_count":19},{"model":"Claude Sonnet 4 (Non-reasoning)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-05-22","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 4.0 family deprecated in favor of newer Claude 4.x models.","num_benchmarks":49,"benchmark_group_count":30,"benchmark_coverage_score":16.7,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":2.63,"source_primary":"artificialanalysis.ai","source_primary_share":0.75,"source_primary_group_share":0.6,"capability_breadth":4.17,"capability_primary":"code","capability_primary_share":0.3542,"capability_primary_group_count":8,"capability_non_primary_group_count":22},{"model":"Ling-2.6-1T","provider":"InclusionAI","country":"CN","openness":"open","release_date":"2026-04-23","param_count":1000.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":49,"benchmark_group_count":29,"benchmark_coverage_score":17.0,"source_family_count":12,"source_group_family_count":11,"source_effective_breadth":3.7,"source_primary":"artificialanalysis.ai","source_primary_share":0.6458,"source_primary_group_share":0.4828,"capability_breadth":4.25,"capability_primary":"code","capability_primary_share":0.3958,"capability_primary_group_count":10,"capability_non_primary_group_count":19},{"model":"MiniMax M1 80k","provider":"MiniMax","country":"CN","openness":"open","release_date":"2025-06-01","param_count":456.0,"active_param_count":45.9,"training_compute_flop":4.3240062e+24,"num_benchmarks":49,"benchmark_group_count":32,"benchmark_coverage_score":19.09,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":2.75,"source_primary":"artificialanalysis.ai","source_primary_share":0.7143,"source_primary_group_share":0.5625,"capability_breadth":4.77,"capability_primary":"code","capability_primary_share":0.3061,"capability_primary_group_count":7,"capability_non_primary_group_count":25},{"model":"Llama-3.1-Tulu-3-8B","provider":"Meta","country":"US","openness":"open","release_date":"2024-11-20","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":49,"benchmark_group_count":29,"benchmark_coverage_score":9.28,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.92,"source_primary":"euroeval.com","source_primary_share":0.7551,"source_primary_group_share":0.6897,"capability_breadth":2.32,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6122,"capability_primary_group_count":17,"capability_non_primary_group_count":12},{"model":"DeepSeek V4 Pro (Max)","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2026-04-23","param_count":1600.0,"active_param_count":49.0,"training_compute_flop":9.702e+24,"num_benchmarks":49,"benchmark_group_count":46,"benchmark_coverage_score":18.5,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":5.21,"source_primary":"benchmarklist.com","source_primary_share":0.2653,"source_primary_group_share":0.2826,"capability_breadth":4.63,"capability_primary":"math","capability_primary_share":0.3878,"capability_primary_group_count":17,"capability_non_primary_group_count":29},{"model":"GLM-4 9B Chat","provider":"Zhipu","country":"CN","openness":"open","release_date":"2024-06-05","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":48,"benchmark_group_count":37,"benchmark_coverage_score":14.14,"source_family_count":15,"source_group_family_count":15,"source_effective_breadth":7.24,"source_primary":"huggingface.co/spaces/ai-forever/libra-leaderboard","source_primary_share":0.1915,"source_primary_group_share":0.2432,"capability_breadth":3.53,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4681,"capability_primary_group_count":13,"capability_non_primary_group_count":24},{"model":"Phi-3-small-8k-instruct","provider":"Microsoft","country":"US","openness":"open","release_date":"2024-05-21","param_count":7.4,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":48,"benchmark_group_count":42,"benchmark_coverage_score":21.94,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":4.2,"source_primary":"livebench.ai","source_primary_share":0.3542,"source_primary_group_share":0.4048,"capability_breadth":5.49,"capability_primary":"reasoning_logical","capability_primary_share":0.3333,"capability_primary_group_count":15,"capability_non_primary_group_count":27},{"model":"openchat-3.5-0106","provider":"Other","country":"XX","openness":"open","release_date":"2024-01-06","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":48,"benchmark_group_count":41,"benchmark_coverage_score":14.31,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":6.81,"source_primary":"lk.instruct.kr","source_primary_share":0.1875,"source_primary_group_share":0.2195,"capability_breadth":3.58,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4375,"capability_primary_group_count":16,"capability_non_primary_group_count":25},{"model":"GPT-5.6 Sol (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":48,"benchmark_group_count":28,"benchmark_coverage_score":17.07,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":3.16,"source_primary":"artificialanalysis.ai","source_primary_share":0.7292,"source_primary_group_share":0.5357,"capability_breadth":4.27,"capability_primary":"code","capability_primary_share":0.3958,"capability_primary_group_count":10,"capability_non_primary_group_count":18},{"model":"Qwen 3 Max (Thinking)","provider":"Alibaba","country":"CN","openness":"api","release_date":"2026-01-27","param_count":700.0,"active_param_count":null,"training_compute_flop":1.512e+25,"num_benchmarks":48,"benchmark_group_count":31,"benchmark_coverage_score":20.12,"source_family_count":14,"source_group_family_count":14,"source_effective_breadth":4.31,"source_primary":"artificialanalysis.ai","source_primary_share":0.6458,"source_primary_group_share":0.4516,"capability_breadth":5.03,"capability_primary":"code","capability_primary_share":0.3333,"capability_primary_group_count":7,"capability_non_primary_group_count":24},{"model":"GLM-4.6V","provider":"Zhipu","country":"CN","openness":"open","release_date":"2025-12-09","param_count":106.0,"active_param_count":12.0,"training_compute_flop":null,"num_benchmarks":48,"benchmark_group_count":38,"benchmark_coverage_score":9.28,"source_family_count":14,"source_group_family_count":14,"source_effective_breadth":6.33,"source_primary":"fab-ai.org","source_primary_share":0.2889,"source_primary_group_share":0.2895,"capability_breadth":2.32,"capability_primary":"vision_language","capability_primary_share":0.6444,"capability_primary_group_count":22,"capability_non_primary_group_count":16},{"model":"Phi-3-medium-128k-instruct","provider":"Microsoft","country":"US","openness":"open","release_date":"2024-05-21","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":47,"benchmark_group_count":44,"benchmark_coverage_score":33.0,"source_family_count":14,"source_group_family_count":14,"source_effective_breadth":4.61,"source_primary":"livebench.ai","source_primary_share":0.3617,"source_primary_group_share":0.3864,"capability_breadth":7.1,"capability_primary":"reasoning_logical","capability_primary_share":0.234,"capability_primary_group_count":11,"capability_non_primary_group_count":33},{"model":"Llama 3.3 70B","provider":"Meta","country":"US","openness":"open","release_date":"2024-12-06","param_count":70.0,"active_param_count":null,"training_compute_flop":6.8649768e+24,"num_benchmarks":47,"benchmark_group_count":36,"benchmark_coverage_score":28.03,"source_family_count":24,"source_group_family_count":24,"source_effective_breadth":10.98,"source_primary":"huggingface.co/spaces/mmtu-benchmark/mmtu-leaderboard","source_primary_share":0.2444,"source_primary_group_share":0.25,"capability_breadth":7.01,"capability_primary":"knowledge_general","capability_primary_share":0.2667,"capability_primary_group_count":8,"capability_non_primary_group_count":28},{"model":"GPT-5.6 Terra (xHigh)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":47,"benchmark_group_count":27,"benchmark_coverage_score":17.29,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":2.95,"source_primary":"artificialanalysis.ai","source_primary_share":0.7447,"source_primary_group_share":0.5556,"capability_breadth":4.32,"capability_primary":"code","capability_primary_share":0.383,"capability_primary_group_count":9,"capability_non_primary_group_count":18},{"model":"Nex N2 Pro","provider":"Nex AGI","country":"CN","openness":"open","release_date":"2026-06-03","param_count":397.0,"active_param_count":17.0,"training_compute_flop":null,"num_benchmarks":47,"benchmark_group_count":25,"benchmark_coverage_score":15.85,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.51,"source_primary":"artificialanalysis.ai","source_primary_share":0.7333,"source_primary_group_share":0.6,"capability_breadth":3.96,"capability_primary":"code","capability_primary_share":0.4,"capability_primary_group_count":8,"capability_non_primary_group_count":17},{"model":"GPT-5 Codex (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-09-23","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":47,"benchmark_group_count":29,"benchmark_coverage_score":13.57,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.57,"source_primary":"artificialanalysis.ai","source_primary_share":0.7447,"source_primary_group_share":0.5862,"capability_breadth":3.39,"capability_primary":"code","capability_primary_share":0.4255,"capability_primary_group_count":11,"capability_non_primary_group_count":18},{"model":"Jamba 1.7 Mini","provider":"AI21 Labs","country":"IL","openness":"api","release_date":"2025-07-01","param_count":52.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":47,"benchmark_group_count":29,"benchmark_coverage_score":17.56,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.32,"source_primary":"artificialanalysis.ai","source_primary_share":0.7609,"source_primary_group_share":0.6207,"capability_breadth":4.39,"capability_primary":"code","capability_primary_share":0.3261,"capability_primary_group_count":6,"capability_non_primary_group_count":23},{"model":"Qwen2.5-Math-7B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-16","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":47,"benchmark_group_count":29,"benchmark_coverage_score":10.01,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.91,"source_primary":"euroeval.com","source_primary_share":0.8085,"source_primary_group_share":0.6897,"capability_breadth":2.5,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5745,"capability_primary_group_count":17,"capability_non_primary_group_count":12},{"model":"Gemini 3.1 Pro (Preview) (High)","provider":"Google","country":"US","openness":"api","release_date":"2026-02-19","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":47,"benchmark_group_count":40,"benchmark_coverage_score":31.0,"source_family_count":19,"source_group_family_count":19,"source_effective_breadth":8.6,"source_primary":"benchmarklist.com","source_primary_share":0.2727,"source_primary_group_share":0.275,"capability_breadth":6.82,"capability_primary":"reasoning_logical","capability_primary_share":0.2273,"capability_primary_group_count":9,"capability_non_primary_group_count":31},{"model":"Llama-3.1-Tulu-3-8B-DPO","provider":"Meta","country":"US","openness":"open","release_date":"2024-11-20","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":47,"benchmark_group_count":27,"benchmark_coverage_score":8.71,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.67,"source_primary":"euroeval.com","source_primary_share":0.7872,"source_primary_group_share":0.7407,"capability_breadth":2.18,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6383,"capability_primary_group_count":17,"capability_non_primary_group_count":10},{"model":"Grok 4.20 (Reasoning)","provider":"xAI","country":"US","openness":"api","release_date":"2026-02-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":46,"benchmark_group_count":31,"benchmark_coverage_score":20.95,"source_family_count":14,"source_group_family_count":14,"source_effective_breadth":7.81,"source_primary":"fab-ai.org","source_primary_share":0.4565,"source_primary_group_share":0.2581,"capability_breadth":5.24,"capability_primary":"vision_language","capability_primary_share":0.3478,"capability_primary_group_count":10,"capability_non_primary_group_count":21},{"model":"GPT-5.6 Terra (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":46,"benchmark_group_count":26,"benchmark_coverage_score":15.91,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":2.75,"source_primary":"artificialanalysis.ai","source_primary_share":0.7609,"source_primary_group_share":0.5769,"capability_breadth":3.98,"capability_primary":"code","capability_primary_share":0.413,"capability_primary_group_count":10,"capability_non_primary_group_count":16},{"model":"HyperCLOVA X SEED Think (32B)","provider":"Naver","country":"KR","openness":"api","release_date":"2025-04-24","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":46,"benchmark_group_count":19,"benchmark_coverage_score":16.15,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.23,"source_primary":"artificialanalysis.ai","source_primary_share":0.9565,"source_primary_group_share":0.8947,"capability_breadth":4.04,"capability_primary":"knowledge_general","capability_primary_share":0.3261,"capability_primary_group_count":6,"capability_non_primary_group_count":13},{"model":"Qwen3 Omni 30B A3B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-09-22","param_count":30.0,"active_param_count":3.0,"training_compute_flop":null,"num_benchmarks":46,"benchmark_group_count":28,"benchmark_coverage_score":19.15,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.28,"source_primary":"artificialanalysis.ai","source_primary_share":0.7609,"source_primary_group_share":0.6071,"capability_breadth":4.79,"capability_primary":"code","capability_primary_share":0.3261,"capability_primary_group_count":6,"capability_non_primary_group_count":22},{"model":"Qwen 3 0.6B (Thinking)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":0.6,"active_param_count":null,"training_compute_flop":1.296e+23,"num_benchmarks":46,"benchmark_group_count":25,"benchmark_coverage_score":13.8,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.82,"source_primary":"artificialanalysis.ai","source_primary_share":0.7778,"source_primary_group_share":0.72,"capability_breadth":3.45,"capability_primary":"knowledge_general","capability_primary_share":0.4,"capability_primary_group_count":7,"capability_non_primary_group_count":18},{"model":"Mistral Small 4","provider":"Mistral","country":"FR","openness":"open","release_date":"2026-03-16","param_count":119.0,"active_param_count":6.5,"training_compute_flop":null,"num_benchmarks":46,"benchmark_group_count":32,"benchmark_coverage_score":24.2,"source_family_count":19,"source_group_family_count":18,"source_effective_breadth":8.53,"source_primary":"huggingface.co","source_primary_share":0.2045,"source_primary_group_share":0.2812,"capability_breadth":6.05,"capability_primary":"knowledge_general","capability_primary_share":0.2727,"capability_primary_group_count":6,"capability_non_primary_group_count":26},{"model":"falcon-11B","provider":"TII","country":"UAE","openness":"open","release_date":"2024-05-14","param_count":11.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":46,"benchmark_group_count":28,"benchmark_coverage_score":10.66,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.53,"source_primary":"euroeval.com","source_primary_share":0.6739,"source_primary_group_share":0.5714,"capability_breadth":2.66,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5435,"capability_primary_group_count":15,"capability_non_primary_group_count":13},{"model":"pythia-6.9B","provider":"EleutherAI","country":"US","openness":"open","release_date":"2023-02-14","param_count":6.9,"active_param_count":null,"training_compute_flop":1.242e+22,"num_benchmarks":45,"benchmark_group_count":40,"benchmark_coverage_score":21.49,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.76,"source_primary":"crfm.stanford.edu","source_primary_share":0.4889,"source_primary_group_share":0.55,"capability_breadth":5.37,"capability_primary":"knowledge_general","capability_primary_share":0.2889,"capability_primary_group_count":13,"capability_non_primary_group_count":27},{"model":"Gemini 2.0 Pro (Preview 02-05)","provider":"Google","country":"US","openness":"api","release_date":"2025-02-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 2.0 family deprecated in favor of newer Gemini models.","num_benchmarks":45,"benchmark_group_count":43,"benchmark_coverage_score":20.1,"source_family_count":15,"source_group_family_count":15,"source_effective_breadth":7.55,"source_primary":"huggingface.co/spaces/maum-ai/koffvqa-leaderboard","source_primary_share":0.2444,"source_primary_group_share":0.2558,"capability_breadth":5.02,"capability_primary":"vision_language","capability_primary_share":0.3556,"capability_primary_group_count":16,"capability_non_primary_group_count":27},{"model":"GPT-5.4 (Non-reasoning)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-03-06","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":45,"benchmark_group_count":26,"benchmark_coverage_score":16.98,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":3.16,"source_primary":"artificialanalysis.ai","source_primary_share":0.7273,"source_primary_group_share":0.5385,"capability_breadth":4.25,"capability_primary":"code","capability_primary_share":0.3636,"capability_primary_group_count":7,"capability_non_primary_group_count":19},{"model":"Gemini 3.5 Flash Lite","provider":"Google","country":"US","openness":"api","release_date":null,"param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":45,"benchmark_group_count":26,"benchmark_coverage_score":16.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":2.94,"source_primary":"artificialanalysis.ai","source_primary_share":0.7333,"source_primary_group_share":0.5385,"capability_breadth":4.0,"capability_primary":"code","capability_primary_share":0.3636,"capability_primary_group_count":7,"capability_non_primary_group_count":18},{"model":"Solar Open 100B (Reasoning)","provider":"Upstage","country":"KR","openness":"open","release_date":"2025-12-30","param_count":100.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":45,"benchmark_group_count":17,"benchmark_coverage_score":16.5,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.26,"source_primary":"artificialanalysis.ai","source_primary_share":0.9556,"source_primary_group_share":0.8824,"capability_breadth":4.12,"capability_primary":"knowledge_general","capability_primary_share":0.3111,"capability_primary_group_count":5,"capability_non_primary_group_count":12},{"model":"Mi:dm K 2.5 Pro Preview","provider":"KT","country":"KR","openness":"api","release_date":"2025-12-11","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":45,"benchmark_group_count":19,"benchmark_coverage_score":16.84,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.23,"source_primary":"artificialanalysis.ai","source_primary_share":0.9556,"source_primary_group_share":0.8947,"capability_breadth":4.21,"capability_primary":"knowledge_general","capability_primary_share":0.3111,"capability_primary_group_count":6,"capability_non_primary_group_count":13},{"model":"Viking-13B","provider":"Other","country":"XX","openness":"open","release_date":"2024-02-16","param_count":13.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":45,"benchmark_group_count":25,"benchmark_coverage_score":7.3,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.47,"source_primary":"euroeval.com","source_primary_share":0.8444,"source_primary_group_share":0.8,"capability_breadth":1.83,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7111,"capability_primary_group_count":20,"capability_non_primary_group_count":5},{"model":"WizardLM-2 8x22B","provider":"Microsoft","country":"US","openness":"open","release_date":"2024-04-15","param_count":176.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":44,"benchmark_group_count":37,"benchmark_coverage_score":22.34,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":7.48,"source_primary":"lk.instruct.kr","source_primary_share":0.2093,"source_primary_group_share":0.1892,"capability_breadth":5.59,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3023,"capability_primary_group_count":9,"capability_non_primary_group_count":28},{"model":"Claude Opus 4.6 (Max)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-02-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":44,"benchmark_group_count":43,"benchmark_coverage_score":34.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":2.47,"source_primary":"benchmarklist.com","source_primary_share":0.6047,"source_primary_group_share":0.6047,"capability_breadth":6.4,"capability_primary":"vision_language","capability_primary_share":0.2093,"capability_primary_group_count":9,"capability_non_primary_group_count":34},{"model":"GPT-3.5 Turbo (0301)","provider":"OpenAI","country":"US","openness":"api","release_date":"2023-03-01","param_count":175.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI GPT-3/GPT-3.5 family retired or superseded by newer API models.","num_benchmarks":44,"benchmark_group_count":40,"benchmark_coverage_score":20.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":2.8,"source_primary":"crfm.stanford.edu","source_primary_share":0.55,"source_primary_group_share":0.55,"capability_breadth":5.0,"capability_primary":"knowledge_general","capability_primary_share":0.3,"capability_primary_group_count":12,"capability_non_primary_group_count":28},{"model":"text-curie-001","provider":"OpenAI","country":"US","openness":"api","release_date":"2022-01-27","param_count":6.7,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI completions family retired or superseded by newer API models.","num_benchmarks":44,"benchmark_group_count":37,"benchmark_coverage_score":18.01,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.86,"source_primary":"crfm.stanford.edu","source_primary_share":0.6818,"source_primary_group_share":0.7027,"capability_breadth":4.5,"capability_primary":"knowledge_general","capability_primary_share":0.3409,"capability_primary_group_count":12,"capability_non_primary_group_count":25},{"model":"Llama 3 8B Ita","provider":"Meta","country":"US","openness":"open","release_date":"2024-04-18","param_count":8.0,"active_param_count":null,"training_compute_flop":7.2e+23,"num_benchmarks":44,"benchmark_group_count":30,"benchmark_coverage_score":11.99,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":4.17,"source_primary":"huggingface.co","source_primary_share":0.25,"source_primary_group_share":0.3667,"capability_breadth":3.0,"capability_primary":"knowledge_general","capability_primary_share":0.4773,"capability_primary_group_count":12,"capability_non_primary_group_count":18},{"model":"GPT-5.3 Codex (xHigh)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-02-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":44,"benchmark_group_count":26,"benchmark_coverage_score":12.95,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":2.84,"source_primary":"artificialanalysis.ai","source_primary_share":0.75,"source_primary_group_share":0.5769,"capability_breadth":3.24,"capability_primary":"code","capability_primary_share":0.4651,"capability_primary_group_count":11,"capability_non_primary_group_count":14},{"model":"Falcon-H1R-7B","provider":"TII","country":"UAE","openness":"open","release_date":"2026-01-06","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":44,"benchmark_group_count":25,"benchmark_coverage_score":14.61,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.0,"source_primary":"artificialanalysis.ai","source_primary_share":0.7727,"source_primary_group_share":0.68,"capability_breadth":3.65,"capability_primary":"code","capability_primary_share":0.3409,"capability_primary_group_count":8,"capability_non_primary_group_count":17},{"model":"Gemma 3n E4B Instruct","provider":"Google","country":"US","openness":"open","release_date":"2025-06-26","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":44,"benchmark_group_count":23,"benchmark_coverage_score":16.41,"source_family_count":4,"source_group_family_count":3,"source_effective_breadth":1.43,"source_primary":"artificialanalysis.ai","source_primary_share":0.8409,"source_primary_group_share":0.8261,"capability_breadth":4.1,"capability_primary":"code","capability_primary_share":0.3636,"capability_primary_group_count":7,"capability_non_primary_group_count":16},{"model":"Devstral Medium","provider":"Mistral","country":"FR","openness":"api","release_date":"2025-07-10","param_count":123.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":44,"benchmark_group_count":25,"benchmark_coverage_score":15.43,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":1.85,"source_primary":"artificialanalysis.ai","source_primary_share":0.7955,"source_primary_group_share":0.72,"capability_breadth":3.86,"capability_primary":"code","capability_primary_share":0.4091,"capability_primary_group_count":8,"capability_non_primary_group_count":17},{"model":"Qwen 3 4B (Reasoning)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":4.0,"active_param_count":null,"training_compute_flop":8.64e+23,"num_benchmarks":44,"benchmark_group_count":35,"benchmark_coverage_score":7.33,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.16,"source_primary":"medmarks.ai","source_primary_share":0.6136,"source_primary_group_share":0.6,"capability_breadth":1.83,"capability_primary":"knowledge_general","capability_primary_share":0.7273,"capability_primary_group_count":23,"capability_non_primary_group_count":12},{"model":"Gemini 2.5 Pro (Preview 06-05)","provider":"Google","country":"US","openness":"api","release_date":"2025-06-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 2.5 preview snapshot retired or superseded by newer Gemini models.","num_benchmarks":43,"benchmark_group_count":40,"benchmark_coverage_score":34.48,"source_family_count":17,"source_group_family_count":17,"source_effective_breadth":7.48,"source_primary":"benchmarklist.com","source_primary_share":0.2439,"source_primary_group_share":0.25,"capability_breadth":8.62,"capability_primary":"vision_language","capability_primary_share":0.1951,"capability_primary_group_count":8,"capability_non_primary_group_count":32},{"model":"Mixtral 8x22B","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-04-17","param_count":141.0,"active_param_count":39.0,"training_compute_flop":2.34e+24,"num_benchmarks":43,"benchmark_group_count":30,"benchmark_coverage_score":15.84,"source_family_count":13,"source_group_family_count":12,"source_effective_breadth":5.49,"source_primary":"lk.instruct.kr","source_primary_share":0.225,"source_primary_group_share":0.3,"capability_breadth":3.96,"capability_primary":"knowledge_general","capability_primary_share":0.35,"capability_primary_group_count":11,"capability_non_primary_group_count":19},{"model":"Qwen 2 VL 72B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-19","param_count":72.0,"active_param_count":null,"training_compute_flop":6.048e+23,"num_benchmarks":43,"benchmark_group_count":35,"benchmark_coverage_score":7.6,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":4.8,"source_primary":"huggingface.co/spaces/maum-ai/koffvqa-leaderboard","source_primary_share":0.2558,"source_primary_group_share":0.3143,"capability_breadth":1.9,"capability_primary":"vision_language","capability_primary_share":0.6977,"capability_primary_group_count":26,"capability_non_primary_group_count":9},{"model":"GPT-5.6 Sol (Medium)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":43,"benchmark_group_count":23,"benchmark_coverage_score":17.32,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.2,"source_primary":"artificialanalysis.ai","source_primary_share":0.814,"source_primary_group_share":0.6522,"capability_breadth":4.33,"capability_primary":"code","capability_primary_share":0.3721,"capability_primary_group_count":7,"capability_non_primary_group_count":16},{"model":"GPT-5.6 Luna (xHigh)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":43,"benchmark_group_count":24,"benchmark_coverage_score":15.25,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":3.06,"source_primary":"artificialanalysis.ai","source_primary_share":0.7442,"source_primary_group_share":0.5417,"capability_breadth":3.81,"capability_primary":"code","capability_primary_share":0.4186,"capability_primary_group_count":9,"capability_non_primary_group_count":15},{"model":"GPT-5.6 Luna (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":43,"benchmark_group_count":24,"benchmark_coverage_score":14.36,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":3.03,"source_primary":"artificialanalysis.ai","source_primary_share":0.7442,"source_primary_group_share":0.5417,"capability_breadth":3.59,"capability_primary":"code","capability_primary_share":0.4419,"capability_primary_group_count":10,"capability_non_primary_group_count":14},{"model":"GPT-5.6 Luna (Medium)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":43,"benchmark_group_count":24,"benchmark_coverage_score":15.25,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":3.03,"source_primary":"artificialanalysis.ai","source_primary_share":0.7442,"source_primary_group_share":0.5417,"capability_breadth":3.81,"capability_primary":"code","capability_primary_share":0.4186,"capability_primary_group_count":9,"capability_non_primary_group_count":15},{"model":"Ring-2.6-1T","provider":"InclusionAI","country":"CN","openness":"open","release_date":"2026-05-14","param_count":1000.0,"active_param_count":63.0,"training_compute_flop":null,"num_benchmarks":43,"benchmark_group_count":24,"benchmark_coverage_score":15.91,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":2.4,"source_primary":"artificialanalysis.ai","source_primary_share":0.7907,"source_primary_group_share":0.625,"capability_breadth":3.98,"capability_primary":"code","capability_primary_share":0.3953,"capability_primary_group_count":8,"capability_non_primary_group_count":16},{"model":"Gemma 4 31B (Reasoning)","provider":"Google","country":"US","openness":"open","release_date":"2026-04-02","param_count":30.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":43,"benchmark_group_count":23,"benchmark_coverage_score":19.28,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":2.01,"source_primary":"artificialanalysis.ai","source_primary_share":0.8372,"source_primary_group_share":0.6957,"capability_breadth":4.82,"capability_primary":"code","capability_primary_share":0.3333,"capability_primary_group_count":5,"capability_non_primary_group_count":17},{"model":"ERNIE 5.0 Thinking Preview","provider":"Baidu","country":"CN","openness":"api","release_date":"2026-01-01","param_count":2400.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":43,"benchmark_group_count":20,"benchmark_coverage_score":15.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.34,"source_primary":"artificialanalysis.ai","source_primary_share":0.9302,"source_primary_group_share":0.85,"capability_breadth":3.81,"capability_primary":"knowledge_general","capability_primary_share":0.3488,"capability_primary_group_count":6,"capability_non_primary_group_count":14},{"model":"Solar Pro 3","provider":"Upstage","country":"KR","openness":"api","release_date":"2026-01-27","param_count":102.0,"active_param_count":12.0,"training_compute_flop":null,"num_benchmarks":43,"benchmark_group_count":23,"benchmark_coverage_score":15.39,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.95,"source_primary":"artificialanalysis.ai","source_primary_share":0.8333,"source_primary_group_share":0.6957,"capability_breadth":3.85,"capability_primary":"code","capability_primary_share":0.3659,"capability_primary_group_count":6,"capability_non_primary_group_count":16},{"model":"Step3 VL 10B","provider":"StepFun","country":"CN","openness":"open","release_date":"2026-01-20","param_count":10.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":43,"benchmark_group_count":25,"benchmark_coverage_score":18.17,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.51,"source_primary":"artificialanalysis.ai","source_primary_share":0.7442,"source_primary_group_share":0.56,"capability_breadth":4.54,"capability_primary":"code","capability_primary_share":0.3256,"capability_primary_group_count":7,"capability_non_primary_group_count":18},{"model":"Inkling","provider":"Thinking Machines","country":"US","openness":"open","release_date":"2026-07-15","param_count":975.0,"active_param_count":41.0,"training_compute_flop":null,"num_benchmarks":43,"benchmark_group_count":35,"benchmark_coverage_score":15.31,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":2.07,"source_primary":"vals.ai","source_primary_share":0.5814,"source_primary_group_share":0.6857,"capability_breadth":3.83,"capability_primary":"knowledge_general","capability_primary_share":0.4186,"capability_primary_group_count":12,"capability_non_primary_group_count":23},{"model":"c4ai-command-r-plus","provider":"Cohere","country":"CA","openness":"open","release_date":"2024-04-03","param_count":104.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":42,"benchmark_group_count":35,"benchmark_coverage_score":17.88,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":5.21,"source_primary":"huggingface.co/spaces/xufengduan/humanlikeness","source_primary_share":0.25,"source_primary_group_share":0.2857,"capability_breadth":4.47,"capability_primary":"reasoning_logical","capability_primary_share":0.35,"capability_primary_group_count":14,"capability_non_primary_group_count":21},{"model":"Claude 2.1","provider":"Anthropic","country":"US","openness":"api","release_date":"2023-11-21","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 1/2 family retired or superseded by newer Claude models.","num_benchmarks":42,"benchmark_group_count":36,"benchmark_coverage_score":22.47,"source_family_count":17,"source_group_family_count":17,"source_effective_breadth":9.97,"source_primary":"artificialanalysis.ai","source_primary_share":0.1667,"source_primary_group_share":0.1944,"capability_breadth":5.62,"capability_primary":"knowledge_general","capability_primary_share":0.3095,"capability_primary_group_count":11,"capability_non_primary_group_count":25},{"model":"GPT-4 (0314)","provider":"OpenAI","country":"US","openness":"api","release_date":"2023-03-14","param_count":1760.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI GPT-4/GPT-4 Turbo family retired or superseded by newer API models.","num_benchmarks":42,"benchmark_group_count":38,"benchmark_coverage_score":23.02,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":7.15,"source_primary":"huggingface.co/spaces/ai-secure/llm-trustworthy-leaderboard","source_primary_share":0.2,"source_primary_group_share":0.2105,"capability_breadth":5.76,"capability_primary":"instruction_following","capability_primary_share":0.3,"capability_primary_group_count":12,"capability_non_primary_group_count":26},{"model":"GPT-5.6 Terra (Medium)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":42,"benchmark_group_count":23,"benchmark_coverage_score":16.88,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.2,"source_primary":"artificialanalysis.ai","source_primary_share":0.8095,"source_primary_group_share":0.6522,"capability_breadth":4.22,"capability_primary":"code","capability_primary_share":0.381,"capability_primary_group_count":7,"capability_non_primary_group_count":16},{"model":"GPT-5.6 Sol (Low)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":42,"benchmark_group_count":22,"benchmark_coverage_score":16.72,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.02,"source_primary":"artificialanalysis.ai","source_primary_share":0.8333,"source_primary_group_share":0.6818,"capability_breadth":4.18,"capability_primary":"code","capability_primary_share":0.381,"capability_primary_group_count":7,"capability_non_primary_group_count":15},{"model":"GPT-5.6 Terra (Low)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":42,"benchmark_group_count":22,"benchmark_coverage_score":16.72,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.02,"source_primary":"artificialanalysis.ai","source_primary_share":0.8333,"source_primary_group_share":0.6818,"capability_breadth":4.18,"capability_primary":"code","capability_primary_share":0.381,"capability_primary_group_count":7,"capability_non_primary_group_count":15},{"model":"Hy3-preview (Reasoning)","provider":"Tencent","country":"CN","openness":"open","release_date":"2026-04-24","param_count":295.0,"active_param_count":21.0,"training_compute_flop":null,"num_benchmarks":42,"benchmark_group_count":23,"benchmark_coverage_score":15.53,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.37,"source_primary":"artificialanalysis.ai","source_primary_share":0.7561,"source_primary_group_share":0.6087,"capability_breadth":3.88,"capability_primary":"code","capability_primary_share":0.3659,"capability_primary_group_count":6,"capability_non_primary_group_count":17},{"model":"GLM-4.7 (Reasoning)","provider":"Zhipu","country":"CN","openness":"open","release_date":"2025-12-22","param_count":355.0,"active_param_count":32.0,"training_compute_flop":4.42e+24,"num_benchmarks":42,"benchmark_group_count":22,"benchmark_coverage_score":14.65,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.45,"source_primary":"artificialanalysis.ai","source_primary_share":0.9024,"source_primary_group_share":0.8182,"capability_breadth":3.66,"capability_primary":"code","capability_primary_share":0.3659,"capability_primary_group_count":6,"capability_non_primary_group_count":16},{"model":"Gemini 3.5 Flash (Minimal)","provider":"Google","country":"US","openness":"api","release_date":"2026-05-19","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":42,"benchmark_group_count":22,"benchmark_coverage_score":16.77,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.26,"source_primary":"artificialanalysis.ai","source_primary_share":0.7805,"source_primary_group_share":0.6364,"capability_breadth":4.19,"capability_primary":"code","capability_primary_share":0.3415,"capability_primary_group_count":5,"capability_non_primary_group_count":17},{"model":"LFM2 24B A2B","provider":"Liquid AI","country":"US","openness":"open","release_date":"2025-05-29","param_count":24.0,"active_param_count":2.0,"training_compute_flop":null,"num_benchmarks":42,"benchmark_group_count":24,"benchmark_coverage_score":16.05,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.67,"source_primary":"artificialanalysis.ai","source_primary_share":0.7561,"source_primary_group_share":0.5833,"capability_breadth":4.01,"capability_primary":"code","capability_primary_share":0.3659,"capability_primary_group_count":6,"capability_non_primary_group_count":18},{"model":"LFM2.5-VL-1.6B","provider":"Liquid AI","country":"US","openness":"open","release_date":"2025-06-17","param_count":1.6,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":42,"benchmark_group_count":19,"benchmark_coverage_score":16.96,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.73,"source_primary":"artificialanalysis.ai","source_primary_share":0.7619,"source_primary_group_share":0.7368,"capability_breadth":4.24,"capability_primary":"code","capability_primary_share":0.3333,"capability_primary_group_count":5,"capability_non_primary_group_count":14},{"model":"Molmo2-8B","provider":"Allen AI","country":"US","openness":"open","release_date":"2025-03-13","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":42,"benchmark_group_count":23,"benchmark_coverage_score":17.29,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.25,"source_primary":"artificialanalysis.ai","source_primary_share":0.7619,"source_primary_group_share":0.6087,"capability_breadth":4.32,"capability_primary":"code","capability_primary_share":0.3333,"capability_primary_group_count":5,"capability_non_primary_group_count":18},{"model":"Nanbeige4.1-3B","provider":"Nanbeige","country":"CN","openness":"open","release_date":"2026-02-11","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":42,"benchmark_group_count":22,"benchmark_coverage_score":14.59,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.03,"source_primary":"artificialanalysis.ai","source_primary_share":0.8293,"source_primary_group_share":0.6818,"capability_breadth":3.65,"capability_primary":"knowledge_general","capability_primary_share":0.3659,"capability_primary_group_count":7,"capability_non_primary_group_count":15},{"model":"gemma-3n-E2B-it Instruct","provider":"Google","country":"US","openness":"open","release_date":"2025-06-26","param_count":5.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":42,"benchmark_group_count":22,"benchmark_coverage_score":15.14,"source_family_count":4,"source_group_family_count":3,"source_effective_breadth":1.45,"source_primary":"artificialanalysis.ai","source_primary_share":0.8333,"source_primary_group_share":0.8182,"capability_breadth":3.79,"capability_primary":"code","capability_primary_share":0.381,"capability_primary_group_count":7,"capability_non_primary_group_count":15},{"model":"GPT-5.2 (Low)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-12-11","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":41,"benchmark_group_count":35,"benchmark_coverage_score":16.0,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":3.89,"source_primary":"ppbench.com","source_primary_share":0.4722,"source_primary_group_share":0.4571,"capability_breadth":4.0,"capability_primary":"reasoning_logical","capability_primary_share":0.4444,"capability_primary_group_count":15,"capability_non_primary_group_count":20},{"model":"GPT-4 Preview (0125)","provider":"OpenAI","country":"US","openness":"api","release_date":"2024-01-25","param_count":1760.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":41,"benchmark_group_count":39,"benchmark_coverage_score":30.0,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":3.97,"source_primary":"livebench.ai","source_primary_share":0.4146,"source_primary_group_share":0.4359,"capability_breadth":6.86,"capability_primary":"reasoning_logical","capability_primary_share":0.2195,"capability_primary_group_count":9,"capability_non_primary_group_count":30},{"model":"Claude Opus 4.7 (Adaptive Reasoning, Max Effort)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-04-16","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":41,"benchmark_group_count":20,"benchmark_coverage_score":15.17,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.5,"source_primary":"artificialanalysis.ai","source_primary_share":0.9,"source_primary_group_share":0.8,"capability_breadth":3.79,"capability_primary":"code","capability_primary_share":0.359,"capability_primary_group_count":5,"capability_non_primary_group_count":14},{"model":"Command A+","provider":"Cohere","country":"CA","openness":"open","release_date":"2026-05-20","param_count":218.0,"active_param_count":25.0,"training_compute_flop":null,"num_benchmarks":41,"benchmark_group_count":21,"benchmark_coverage_score":17.3,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.62,"source_primary":"artificialanalysis.ai","source_primary_share":0.878,"source_primary_group_share":0.7619,"capability_breadth":4.32,"capability_primary":"code","capability_primary_share":0.35,"capability_primary_group_count":5,"capability_non_primary_group_count":15},{"model":"Claude Haiku 4.5 (Non-reasoning)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-10-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":41,"benchmark_group_count":22,"benchmark_coverage_score":13.62,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.61,"source_primary":"artificialanalysis.ai","source_primary_share":0.875,"source_primary_group_share":0.7727,"capability_breadth":3.4,"capability_primary":"code","capability_primary_share":0.4,"capability_primary_group_count":7,"capability_non_primary_group_count":15},{"model":"Granite 4.1 8B","provider":"IBM","country":"US","openness":"open","release_date":"2026-04-29","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":41,"benchmark_group_count":23,"benchmark_coverage_score":13.91,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.2,"source_primary":"artificialanalysis.ai","source_primary_share":0.8,"source_primary_group_share":0.6522,"capability_breadth":3.48,"capability_primary":"code","capability_primary_share":0.425,"capability_primary_group_count":8,"capability_non_primary_group_count":15},{"model":"KAT Coder Pro V2","provider":"Kuaishou","country":"CN","openness":"api","release_date":"2026-03-27","param_count":null,"active_param_count":72.0,"training_compute_flop":null,"num_benchmarks":41,"benchmark_group_count":21,"benchmark_coverage_score":15.25,"source_family_count":6,"source_group_family_count":5,"source_effective_breadth":1.86,"source_primary":"artificialanalysis.ai","source_primary_share":0.8293,"source_primary_group_share":0.7143,"capability_breadth":3.81,"capability_primary":"code","capability_primary_share":0.3902,"capability_primary_group_count":7,"capability_non_primary_group_count":14},{"model":"Qwen 3.5 2B (Non-reasoning)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-03-02","param_count":2.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":41,"benchmark_group_count":20,"benchmark_coverage_score":15.76,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"artificialanalysis.ai","source_primary_share":0.875,"source_primary_group_share":0.75,"capability_breadth":3.94,"capability_primary":"code","capability_primary_share":0.35,"capability_primary_group_count":5,"capability_non_primary_group_count":15},{"model":"Qwen 3 1.7B (Thinking)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":1.7,"active_param_count":null,"training_compute_flop":3.672e+23,"num_benchmarks":41,"benchmark_group_count":21,"benchmark_coverage_score":13.06,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.34,"source_primary":"artificialanalysis.ai","source_primary_share":0.8537,"source_primary_group_share":0.8571,"capability_breadth":3.26,"capability_primary":"knowledge_general","capability_primary_share":0.3902,"capability_primary_group_count":6,"capability_non_primary_group_count":15},{"model":"Claude Sonnet 4.6 (High)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-02-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":41,"benchmark_group_count":34,"benchmark_coverage_score":26.16,"source_family_count":14,"source_group_family_count":14,"source_effective_breadth":4.86,"source_primary":"rank.opencompass.org.cn","source_primary_share":0.5122,"source_primary_group_share":0.4118,"capability_breadth":6.54,"capability_primary":"knowledge_general","capability_primary_share":0.2927,"capability_primary_group_count":11,"capability_non_primary_group_count":23},{"model":"Gemini 1.5 Pro (0514)","provider":"Google","country":"US","openness":"api","release_date":"2024-02-15","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 1.x family retired or superseded by newer Gemini models.","num_benchmarks":40,"benchmark_group_count":39,"benchmark_coverage_score":30.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.67,"source_primary":"livebench.ai","source_primary_share":0.425,"source_primary_group_share":0.4359,"capability_breadth":7.08,"capability_primary":"reasoning_logical","capability_primary_share":0.225,"capability_primary_group_count":9,"capability_non_primary_group_count":30},{"model":"Mixtral 8x7B Instruct","provider":"Mistral","country":"FR","openness":"open","release_date":"2023-12-11","param_count":46.7,"active_param_count":12.9,"training_compute_flop":7.74e+23,"num_benchmarks":40,"benchmark_group_count":30,"benchmark_coverage_score":16.41,"source_family_count":14,"source_group_family_count":14,"source_effective_breadth":7.26,"source_primary":"lk.instruct.kr","source_primary_share":0.225,"source_primary_group_share":0.2333,"capability_breadth":4.1,"capability_primary":"code","capability_primary_share":0.375,"capability_primary_group_count":10,"capability_non_primary_group_count":20},{"model":"Llama 3 8B Cpt Sea Lionv2.1 Instruct","provider":"Meta","country":"US","openness":"open","release_date":null,"param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":40,"benchmark_group_count":27,"benchmark_coverage_score":9.5,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.36,"source_primary":"huggingface.co/spaces/meralion/seaeval_leaderboard","source_primary_share":0.85,"source_primary_group_share":0.8519,"capability_breadth":2.37,"capability_primary":"knowledge_multilingual","capability_primary_share":0.575,"capability_primary_group_count":14,"capability_non_primary_group_count":13},{"model":"Claude Opus 4.8 (Adaptive Reasoning, Max Effort)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-05-29","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":40,"benchmark_group_count":20,"benchmark_coverage_score":14.51,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.5,"source_primary":"artificialanalysis.ai","source_primary_share":0.8974,"source_primary_group_share":0.8,"capability_breadth":3.63,"capability_primary":"code","capability_primary_share":0.3684,"capability_primary_group_count":5,"capability_non_primary_group_count":14},{"model":"Hy3-preview (Non-reasoning)","provider":"Tencent","country":"CN","openness":"open","release_date":"2026-04-24","param_count":295.0,"active_param_count":21.0,"training_compute_flop":null,"num_benchmarks":40,"benchmark_group_count":21,"benchmark_coverage_score":15.17,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.0,"source_primary":"artificialanalysis.ai","source_primary_share":0.7949,"source_primary_group_share":0.6667,"capability_breadth":3.79,"capability_primary":"code","capability_primary_share":0.359,"capability_primary_group_count":5,"capability_non_primary_group_count":16},{"model":"Nova 2.0 Pro Preview (Medium)","provider":"Amazon","country":"US","openness":"api","release_date":"2025-08-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":40,"benchmark_group_count":20,"benchmark_coverage_score":15.38,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.22,"source_primary":"artificialanalysis.ai","source_primary_share":0.95,"source_primary_group_share":0.9,"capability_breadth":3.85,"capability_primary":"code","capability_primary_share":0.375,"capability_primary_group_count":6,"capability_non_primary_group_count":14},{"model":"Granite 4.0 1B","provider":"IBM","country":"US","openness":"open","release_date":"2025-10-01","param_count":1.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":40,"benchmark_group_count":20,"benchmark_coverage_score":15.61,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.36,"source_primary":"artificialanalysis.ai","source_primary_share":0.85,"source_primary_group_share":0.85,"capability_breadth":3.9,"capability_primary":"code","capability_primary_share":0.375,"capability_primary_group_count":6,"capability_non_primary_group_count":14},{"model":"Qwen 3 VL 235B A22B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-09-22","param_count":235.0,"active_param_count":22.0,"training_compute_flop":null,"num_benchmarks":40,"benchmark_group_count":35,"benchmark_coverage_score":11.85,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":5.0,"source_primary":"benchmarklist.com","source_primary_share":0.3,"source_primary_group_share":0.3429,"capability_breadth":2.96,"capability_primary":"vision_language","capability_primary_share":0.525,"capability_primary_group_count":21,"capability_non_primary_group_count":14},{"model":"Claude Opus 4.5 (High)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-11-24","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":40,"benchmark_group_count":36,"benchmark_coverage_score":17.61,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":3.93,"source_primary":"ppbench.com","source_primary_share":0.4595,"source_primary_group_share":0.4444,"capability_breadth":4.4,"capability_primary":"reasoning_logical","capability_primary_share":0.4054,"capability_primary_group_count":14,"capability_non_primary_group_count":22},{"model":"GPT-OSS-120B (Medium)","provider":"OpenAI","country":"US","openness":"open","release_date":"2025-08-05","param_count":120.0,"active_param_count":5.13,"training_compute_flop":4.94e+24,"num_benchmarks":40,"benchmark_group_count":33,"benchmark_coverage_score":8.96,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":2.34,"source_primary":"medmarks.ai","source_primary_share":0.6923,"source_primary_group_share":0.6364,"capability_breadth":2.24,"capability_primary":"knowledge_general","capability_primary_share":0.641,"capability_primary_group_count":19,"capability_non_primary_group_count":14},{"model":"text-davinci-002","provider":"OpenAI","country":"US","openness":"api","release_date":"2022-03-15","param_count":175.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI completions family retired or superseded by newer API models.","num_benchmarks":39,"benchmark_group_count":35,"benchmark_coverage_score":13.67,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.7,"source_primary":"crfm.stanford.edu","source_primary_share":0.7692,"source_primary_group_share":0.7429,"capability_breadth":3.42,"capability_primary":"knowledge_general","capability_primary_share":0.4615,"capability_primary_group_count":15,"capability_non_primary_group_count":20},{"model":"GPT-5.6 Luna (Low)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":39,"benchmark_group_count":20,"benchmark_coverage_score":15.02,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.17,"source_primary":"artificialanalysis.ai","source_primary_share":0.8205,"source_primary_group_share":0.65,"capability_breadth":3.76,"capability_primary":"code","capability_primary_share":0.4103,"capability_primary_group_count":7,"capability_non_primary_group_count":13},{"model":"DeepSeek V3.2 Exp (Non-reasoning)","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-09-29","param_count":671.0,"active_param_count":37.0,"training_compute_flop":4.18e+24,"num_benchmarks":39,"benchmark_group_count":21,"benchmark_coverage_score":13.19,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.47,"source_primary":"artificialanalysis.ai","source_primary_share":0.8947,"source_primary_group_share":0.8095,"capability_breadth":3.3,"capability_primary":"code","capability_primary_share":0.3947,"capability_primary_group_count":6,"capability_non_primary_group_count":15},{"model":"Trinity Large (Thinking)","provider":"Arcee AI","country":"US","openness":"open","release_date":"2026-04-01","param_count":400.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":39,"benchmark_group_count":20,"benchmark_coverage_score":14.8,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.69,"source_primary":"artificialanalysis.ai","source_primary_share":0.8718,"source_primary_group_share":0.75,"capability_breadth":3.7,"capability_primary":"code","capability_primary_share":0.3846,"capability_primary_group_count":6,"capability_non_primary_group_count":14},{"model":"Qwen 3.6 27B (Non-reasoning)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-04-22","param_count":27.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":39,"benchmark_group_count":18,"benchmark_coverage_score":14.51,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.38,"source_primary":"artificialanalysis.ai","source_primary_share":0.9211,"source_primary_group_share":0.8333,"capability_breadth":3.63,"capability_primary":"code","capability_primary_share":0.3684,"capability_primary_group_count":5,"capability_non_primary_group_count":13},{"model":"Qwen 3.5 Omni Plus","provider":"Alibaba","country":"CN","openness":"api","release_date":"2026-03-30","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":39,"benchmark_group_count":21,"benchmark_coverage_score":16.4,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.98,"source_primary":"artificialanalysis.ai","source_primary_share":0.8205,"source_primary_group_share":0.6667,"capability_breadth":4.1,"capability_primary":"code","capability_primary_share":0.359,"capability_primary_group_count":5,"capability_non_primary_group_count":16},{"model":"Granite 4.0 350M","provider":"IBM","country":"US","openness":"open","release_date":"2025-10-01","param_count":0.35,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":39,"benchmark_group_count":22,"benchmark_coverage_score":14.88,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.47,"source_primary":"artificialanalysis.ai","source_primary_share":0.8974,"source_primary_group_share":0.8182,"capability_breadth":3.72,"capability_primary":"code","capability_primary_share":0.3846,"capability_primary_group_count":6,"capability_non_primary_group_count":16},{"model":"Qwen 3.6 35B A3B (Non-reasoning)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-04-16","param_count":35.0,"active_param_count":3.0,"training_compute_flop":null,"num_benchmarks":39,"benchmark_group_count":18,"benchmark_coverage_score":14.51,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.38,"source_primary":"artificialanalysis.ai","source_primary_share":0.9211,"source_primary_group_share":0.8333,"capability_breadth":3.63,"capability_primary":"code","capability_primary_share":0.3684,"capability_primary_group_count":5,"capability_non_primary_group_count":13},{"model":"Qwen 3.5 4B (Non-reasoning)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-03-01","param_count":4.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":39,"benchmark_group_count":20,"benchmark_coverage_score":14.81,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"artificialanalysis.ai","source_primary_share":0.8684,"source_primary_group_share":0.75,"capability_breadth":3.7,"capability_primary":"code","capability_primary_share":0.3684,"capability_primary_group_count":5,"capability_non_primary_group_count":15},{"model":"DeepSeek V3.1 (Non-reasoning)","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-08-21","param_count":671.0,"active_param_count":37.0,"training_compute_flop":3.594058e+24,"num_benchmarks":39,"benchmark_group_count":21,"benchmark_coverage_score":13.19,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.47,"source_primary":"artificialanalysis.ai","source_primary_share":0.8947,"source_primary_group_share":0.8095,"capability_breadth":3.3,"capability_primary":"code","capability_primary_share":0.3947,"capability_primary_group_count":6,"capability_non_primary_group_count":15},{"model":"GLM-4.5 (Reasoning)","provider":"Zhipu","country":"CN","openness":"open","release_date":"2025-07-28","param_count":355.0,"active_param_count":32.0,"training_compute_flop":4.42e+24,"num_benchmarks":39,"benchmark_group_count":21,"benchmark_coverage_score":13.01,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"artificialanalysis.ai","source_primary_share":0.9211,"source_primary_group_share":0.8571,"capability_breadth":3.25,"capability_primary":"code","capability_primary_share":0.3947,"capability_primary_group_count":6,"capability_non_primary_group_count":15},{"model":"GLM-4.6 (Non-reasoning)","provider":"Zhipu","country":"CN","openness":"open","release_date":"2025-09-30","param_count":357.0,"active_param_count":32.0,"training_compute_flop":4.42e+24,"num_benchmarks":39,"benchmark_group_count":21,"benchmark_coverage_score":13.19,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.47,"source_primary":"artificialanalysis.ai","source_primary_share":0.8947,"source_primary_group_share":0.8095,"capability_breadth":3.3,"capability_primary":"code","capability_primary_share":0.3947,"capability_primary_group_count":6,"capability_non_primary_group_count":15},{"model":"Gemini 3 Pro (High)","provider":"Google","country":"US","openness":"api","release_date":"2025-11-18","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":39,"benchmark_group_count":34,"benchmark_coverage_score":15.08,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.4,"source_primary":"ppbench.com","source_primary_share":0.4857,"source_primary_group_share":0.4706,"capability_breadth":3.77,"capability_primary":"reasoning_logical","capability_primary_share":0.4571,"capability_primary_group_count":15,"capability_non_primary_group_count":19},{"model":"GPT-5.4 (Medium)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-03-06","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":39,"benchmark_group_count":26,"benchmark_coverage_score":16.14,"source_family_count":15,"source_group_family_count":15,"source_effective_breadth":7.51,"source_primary":"fab-ai.org","source_primary_share":0.5385,"source_primary_group_share":0.3077,"capability_breadth":4.03,"capability_primary":"vision_language","capability_primary_share":0.4103,"capability_primary_group_count":10,"capability_non_primary_group_count":16},{"model":"GPT-5.1 Codex","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-11-19","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":39,"benchmark_group_count":36,"benchmark_coverage_score":17.38,"source_family_count":21,"source_group_family_count":21,"source_effective_breadth":12.23,"source_primary":"benchmarklist.com","source_primary_share":0.1892,"source_primary_group_share":0.1944,"capability_breadth":4.35,"capability_primary":"code","capability_primary_share":0.4054,"capability_primary_group_count":14,"capability_non_primary_group_count":22},{"model":"Claude Opus 4.7 (High)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-04-16","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":39,"benchmark_group_count":31,"benchmark_coverage_score":26.8,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":4.2,"source_primary":"rank.opencompass.org.cn","source_primary_share":0.5385,"source_primary_group_share":0.4516,"capability_breadth":6.7,"capability_primary":"code","capability_primary_share":0.2308,"capability_primary_group_count":8,"capability_non_primary_group_count":23},{"model":"Pixtral Large","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-11-18","param_count":124.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":38,"benchmark_group_count":31,"benchmark_coverage_score":9.41,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":4.43,"source_primary":"fab-ai.org","source_primary_share":0.3421,"source_primary_group_share":0.3548,"capability_breadth":2.35,"capability_primary":"vision_language","capability_primary_share":0.6316,"capability_primary_group_count":17,"capability_non_primary_group_count":14},{"model":"Mistral-Small-Instruct-2409","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-09-17","param_count":22.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":38,"benchmark_group_count":30,"benchmark_coverage_score":16.75,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":5.7,"source_primary":"huggingface.co/spaces/speakleash/mt-bench-pl","source_primary_share":0.2432,"source_primary_group_share":0.3,"capability_breadth":4.19,"capability_primary":"knowledge_general","capability_primary_share":0.3243,"capability_primary_group_count":9,"capability_non_primary_group_count":21},{"model":"OpenHermes-2.5-Mistral-7B","provider":"Other","country":"XX","openness":"open","release_date":"2023-10-29","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":38,"benchmark_group_count":34,"benchmark_coverage_score":25.33,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":3.25,"source_primary":"livebench.ai","source_primary_share":0.4474,"source_primary_group_share":0.5,"capability_breadth":6.33,"capability_primary":"reasoning_logical","capability_primary_share":0.2895,"capability_primary_group_count":10,"capability_non_primary_group_count":24},{"model":"Hermes 4 405B (Thinking)","provider":"Nous Research","country":"US","openness":"open","release_date":"2025-08-26","param_count":405.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":38,"benchmark_group_count":21,"benchmark_coverage_score":14.51,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.49,"source_primary":"artificialanalysis.ai","source_primary_share":0.8947,"source_primary_group_share":0.8095,"capability_breadth":3.63,"capability_primary":"code","capability_primary_share":0.3947,"capability_primary_group_count":6,"capability_non_primary_group_count":15},{"model":"Ring-flash-2.0","provider":"InclusionAI","country":"CN","openness":"open","release_date":"2025-09-19","param_count":100.0,"active_param_count":6.1,"training_compute_flop":null,"num_benchmarks":38,"benchmark_group_count":21,"benchmark_coverage_score":13.75,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.47,"source_primary":"artificialanalysis.ai","source_primary_share":0.8947,"source_primary_group_share":0.8095,"capability_breadth":3.44,"capability_primary":"code","capability_primary_share":0.3947,"capability_primary_group_count":6,"capability_non_primary_group_count":15},{"model":"Granite 4.0 H 350M","provider":"IBM","country":"US","openness":"open","release_date":"2025-10-01","param_count":0.35,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":38,"benchmark_group_count":21,"benchmark_coverage_score":14.3,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.34,"source_primary":"artificialanalysis.ai","source_primary_share":0.9211,"source_primary_group_share":0.8571,"capability_breadth":3.57,"capability_primary":"code","capability_primary_share":0.3947,"capability_primary_group_count":6,"capability_non_primary_group_count":15},{"model":"LongCat Flash Lite","provider":"Meituan","country":"CN","openness":"open","release_date":"2026-01-29","param_count":68.5,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":38,"benchmark_group_count":21,"benchmark_coverage_score":13.56,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.11,"source_primary":"artificialanalysis.ai","source_primary_share":0.8158,"source_primary_group_share":0.6667,"capability_breadth":3.39,"capability_primary":"knowledge_general","capability_primary_share":0.3684,"capability_primary_group_count":6,"capability_non_primary_group_count":15},{"model":"Qwen 3.5 0.8B (Thinking)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-03-02","param_count":0.8,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":38,"benchmark_group_count":18,"benchmark_coverage_score":15.87,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.25,"source_primary":"artificialanalysis.ai","source_primary_share":0.9474,"source_primary_group_share":0.8889,"capability_breadth":3.97,"capability_primary":"code","capability_primary_share":0.3684,"capability_primary_group_count":5,"capability_non_primary_group_count":13},{"model":"Claude 3.7 Sonnet (Non-reasoning)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-02-24","param_count":null,"active_param_count":null,"training_compute_flop":3.35e+25,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 3 family retired or superseded by newer Claude models.","num_benchmarks":38,"benchmark_group_count":20,"benchmark_coverage_score":13.62,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.23,"source_primary":"artificialanalysis.ai","source_primary_share":0.9474,"source_primary_group_share":0.9,"capability_breadth":3.41,"capability_primary":"code","capability_primary_share":0.4211,"capability_primary_group_count":7,"capability_non_primary_group_count":13},{"model":"Claude Opus 4.5 (20251101) (Thinking)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-11-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":38,"benchmark_group_count":35,"benchmark_coverage_score":15.25,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.22,"source_primary":"vals.ai","source_primary_share":0.5833,"source_primary_group_share":0.6,"capability_breadth":3.81,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":15,"capability_non_primary_group_count":20},{"model":"granite-3.0-3B-a800m-base","provider":"IBM","country":"US","openness":"open","release_date":"2024-10-03","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":38,"benchmark_group_count":23,"benchmark_coverage_score":10.31,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.81,"source_primary":"euroeval.com","source_primary_share":0.8158,"source_primary_group_share":0.6957,"capability_breadth":2.58,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5526,"capability_primary_group_count":14,"capability_non_primary_group_count":9},{"model":"O3 Mini (Medium)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-01-31","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":37,"benchmark_group_count":36,"benchmark_coverage_score":31.29,"source_family_count":22,"source_group_family_count":22,"source_effective_breadth":11.37,"source_primary":"prollm.ai","source_primary_share":0.2162,"source_primary_group_share":0.2222,"capability_breadth":7.82,"capability_primary":"reasoning_logical","capability_primary_share":0.1892,"capability_primary_group_count":7,"capability_non_primary_group_count":29},{"model":"granite-3.0-2B-instruct","provider":"IBM","country":"US","openness":"open","release_date":"2024-10-02","param_count":2.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":37,"benchmark_group_count":29,"benchmark_coverage_score":12.59,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.81,"source_primary":"euroeval.com","source_primary_share":0.4054,"source_primary_group_share":0.3448,"capability_breadth":3.15,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4865,"capability_primary_group_count":13,"capability_non_primary_group_count":16},{"model":"Qwen-14B-Chat","provider":"Alibaba","country":"CN","openness":"open","release_date":"2023-09-25","param_count":14.0,"active_param_count":null,"training_compute_flop":2.5e+23,"num_benchmarks":37,"benchmark_group_count":30,"benchmark_coverage_score":29.45,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":5.92,"source_primary":"huggingface.co/spaces/nphardeval/nphardeval-leaderboard","source_primary_share":0.2778,"source_primary_group_share":0.2667,"capability_breadth":7.36,"capability_primary":"reasoning_logical","capability_primary_share":0.1944,"capability_primary_group_count":6,"capability_non_primary_group_count":24},{"model":"Claude Sonnet 5 (Adaptive Reasoning, Max Effort)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-06-30","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":37,"benchmark_group_count":17,"benchmark_coverage_score":12.86,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.41,"source_primary":"artificialanalysis.ai","source_primary_share":0.9167,"source_primary_group_share":0.8235,"capability_breadth":3.22,"capability_primary":"code","capability_primary_share":0.4,"capability_primary_group_count":5,"capability_non_primary_group_count":11},{"model":"GPT-5.2 Codex (xHigh)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-01-14","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":37,"benchmark_group_count":19,"benchmark_coverage_score":12.82,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":1.8,"source_primary":"artificialanalysis.ai","source_primary_share":0.8649,"source_primary_group_share":0.7368,"capability_breadth":3.21,"capability_primary":"code","capability_primary_share":0.4595,"capability_primary_group_count":8,"capability_non_primary_group_count":11},{"model":"Qwen 3.5 9B (Non-reasoning)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-03-02","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":37,"benchmark_group_count":18,"benchmark_coverage_score":13.57,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.38,"source_primary":"artificialanalysis.ai","source_primary_share":0.9167,"source_primary_group_share":0.8333,"capability_breadth":3.39,"capability_primary":"code","capability_primary_share":0.3889,"capability_primary_group_count":5,"capability_non_primary_group_count":13},{"model":"GPT-5.6 Luna (Non-reasoning)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":37,"benchmark_group_count":17,"benchmark_coverage_score":12.64,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.61,"source_primary":"artificialanalysis.ai","source_primary_share":0.8889,"source_primary_group_share":0.7647,"capability_breadth":3.16,"capability_primary":"code","capability_primary_share":0.4167,"capability_primary_group_count":6,"capability_non_primary_group_count":11},{"model":"Llama 3.1 Nemotron Ultra 253B v1 (Reasoning)","provider":"NVIDIA","country":"US","openness":"open","release_date":"2025-04-07","param_count":253.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":37,"benchmark_group_count":20,"benchmark_coverage_score":13.59,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.22,"source_primary":"artificialanalysis.ai","source_primary_share":0.9459,"source_primary_group_share":0.9,"capability_breadth":3.4,"capability_primary":"code","capability_primary_share":0.4054,"capability_primary_group_count":6,"capability_non_primary_group_count":14},{"model":"Granite 4.1 30B","provider":"IBM","country":"US","openness":"open","release_date":"2026-04-29","param_count":30.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":37,"benchmark_group_count":18,"benchmark_coverage_score":14.22,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.38,"source_primary":"artificialanalysis.ai","source_primary_share":0.9189,"source_primary_group_share":0.8333,"capability_breadth":3.56,"capability_primary":"code","capability_primary_share":0.4054,"capability_primary_group_count":6,"capability_non_primary_group_count":12},{"model":"Hermes 4 70B (Thinking)","provider":"Nous Research","country":"US","openness":"open","release_date":"2025-08-26","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":37,"benchmark_group_count":20,"benchmark_coverage_score":13.79,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.36,"source_primary":"artificialanalysis.ai","source_primary_share":0.9189,"source_primary_group_share":0.85,"capability_breadth":3.45,"capability_primary":"code","capability_primary_share":0.4054,"capability_primary_group_count":6,"capability_non_primary_group_count":14},{"model":"K2 Think V2","provider":"LLM360","country":"UAE","openness":"open","release_date":"2026-01-08","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":37,"benchmark_group_count":18,"benchmark_coverage_score":14.56,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.25,"source_primary":"artificialanalysis.ai","source_primary_share":0.9459,"source_primary_group_share":0.8889,"capability_breadth":3.64,"capability_primary":"code","capability_primary_share":0.3889,"capability_primary_group_count":5,"capability_non_primary_group_count":12},{"model":"MiniCPM-V 4.6 1.3B","provider":"OpenBMB","country":"CN","openness":"open","release_date":"2026-05-11","param_count":1.3,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":37,"benchmark_group_count":17,"benchmark_coverage_score":15.34,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.26,"source_primary":"artificialanalysis.ai","source_primary_share":0.9459,"source_primary_group_share":0.8824,"capability_breadth":3.83,"capability_primary":"code","capability_primary_share":0.3784,"capability_primary_group_count":5,"capability_non_primary_group_count":12},{"model":"Qwen 3.5 0.8B (Non-reasoning)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2026-03-02","param_count":0.8,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":37,"benchmark_group_count":17,"benchmark_coverage_score":15.34,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.26,"source_primary":"artificialanalysis.ai","source_primary_share":0.9459,"source_primary_group_share":0.8824,"capability_breadth":3.83,"capability_primary":"code","capability_primary_share":0.3784,"capability_primary_group_count":5,"capability_non_primary_group_count":12},{"model":"Llama 3.3 Nemotron Super 49B v1 (Reasoning)","provider":"NVIDIA","country":"US","openness":"open","release_date":"2025-03-18","param_count":49.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":37,"benchmark_group_count":20,"benchmark_coverage_score":13.59,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.22,"source_primary":"artificialanalysis.ai","source_primary_share":0.9459,"source_primary_group_share":0.9,"capability_breadth":3.4,"capability_primary":"code","capability_primary_share":0.4054,"capability_primary_group_count":6,"capability_non_primary_group_count":14},{"model":"Granite 3.3 8B (Non-reasoning)","provider":"IBM","country":"US","openness":"open","release_date":"2025-02-26","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":37,"benchmark_group_count":20,"benchmark_coverage_score":13.59,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.22,"source_primary":"artificialanalysis.ai","source_primary_share":0.9459,"source_primary_group_share":0.9,"capability_breadth":3.4,"capability_primary":"code","capability_primary_share":0.4054,"capability_primary_group_count":6,"capability_non_primary_group_count":14},{"model":"GPT-5.2 Codex","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-01-14","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":37,"benchmark_group_count":36,"benchmark_coverage_score":10.09,"source_family_count":24,"source_group_family_count":24,"source_effective_breadth":14.09,"source_primary":"benchmarklist.com","source_primary_share":0.1944,"source_primary_group_share":0.1944,"capability_breadth":2.52,"capability_primary":"code","capability_primary_share":0.6111,"capability_primary_group_count":22,"capability_non_primary_group_count":14},{"model":"Gemini 2.5 Flash (Preview 09-2025)","provider":"Google","country":"US","openness":"api","release_date":"2025-09-25","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 2.5 preview snapshot retired or superseded by newer Gemini models.","num_benchmarks":37,"benchmark_group_count":34,"benchmark_coverage_score":24.78,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":3.83,"source_primary":"vals.ai","source_primary_share":0.3784,"source_primary_group_share":0.4118,"capability_breadth":6.19,"capability_primary":"knowledge_general","capability_primary_share":0.2703,"capability_primary_group_count":10,"capability_non_primary_group_count":24},{"model":"GPT-5.4 (2026-03-05)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-03-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":37,"benchmark_group_count":34,"benchmark_coverage_score":21.25,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":1.77,"source_primary":"vals.ai","source_primary_share":0.6944,"source_primary_group_share":0.7353,"capability_breadth":5.31,"capability_primary":"knowledge_general","capability_primary_share":0.3056,"capability_primary_group_count":11,"capability_non_primary_group_count":23},{"model":"granite-3.0-3B-a800m-instruct","provider":"IBM","country":"US","openness":"open","release_date":"2024-10-03","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":37,"benchmark_group_count":22,"benchmark_coverage_score":10.55,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.66,"source_primary":"euroeval.com","source_primary_share":0.8378,"source_primary_group_share":0.7273,"capability_breadth":2.64,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5405,"capability_primary_group_count":13,"capability_non_primary_group_count":9},{"model":"Claude Opus 4.6 (High)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-02-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":30,"benchmark_coverage_score":10.14,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":3.15,"source_primary":"matharena.ai","source_primary_share":0.6,"source_primary_group_share":0.5333,"capability_breadth":2.54,"capability_primary":"math","capability_primary_share":0.6,"capability_primary_group_count":16,"capability_non_primary_group_count":14},{"model":"O1 Mini (2024-09-12)","provider":"OpenAI","country":"US","openness":"api","release_date":"2024-09-12","param_count":100.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI o1 family deprecated or superseded by newer o-series models.","num_benchmarks":36,"benchmark_group_count":34,"benchmark_coverage_score":25.92,"source_family_count":18,"source_group_family_count":18,"source_effective_breadth":10.32,"source_primary":"huggingface.co/spaces/mideind/icelandic-llm-leaderboard","source_primary_share":0.1944,"source_primary_group_share":0.1765,"capability_breadth":6.48,"capability_primary":"knowledge_multilingual","capability_primary_share":0.25,"capability_primary_group_count":7,"capability_non_primary_group_count":27},{"model":"starchat2-15B-v0.1","provider":"HuggingFace","country":"FR","openness":"open","release_date":"2024-03-10","param_count":15.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":22,"benchmark_coverage_score":9.98,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.23,"source_primary":"huggingface.co/spaces/hpai-bsc/turtle-leaderboard","source_primary_share":0.4,"source_primary_group_share":0.4545,"capability_breadth":2.49,"capability_primary":"code","capability_primary_share":0.5429,"capability_primary_group_count":11,"capability_non_primary_group_count":11},{"model":"DeepSeek V2.5","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2024-09-05","param_count":236.0,"active_param_count":null,"training_compute_flop":1.789e+24,"num_benchmarks":36,"benchmark_group_count":35,"benchmark_coverage_score":21.25,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":3.72,"source_primary":"livebench.ai","source_primary_share":0.4722,"source_primary_group_share":0.4857,"capability_breadth":5.31,"capability_primary":"code","capability_primary_share":0.3056,"capability_primary_group_count":11,"capability_non_primary_group_count":24},{"model":"Gemini 2.5 Flash (Preview 05-20)","provider":"Google","country":"US","openness":"api","release_date":"2025-06-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 2.5 preview snapshot retired or superseded by newer Gemini models.","num_benchmarks":36,"benchmark_group_count":33,"benchmark_coverage_score":15.26,"source_family_count":19,"source_group_family_count":19,"source_effective_breadth":6.94,"source_primary":"huggingface.co/spaces/maum-ai/koffvqa-leaderboard","source_primary_share":0.3143,"source_primary_group_share":0.3333,"capability_breadth":3.82,"capability_primary":"vision_language","capability_primary_share":0.4571,"capability_primary_group_count":16,"capability_non_primary_group_count":17},{"model":"Hermes-3-Llama-3.1-8B","provider":"Nous Research","country":"US","openness":"open","release_date":"2024-07-28","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":24,"benchmark_coverage_score":10.2,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.74,"source_primary":"huggingface.co","source_primary_share":0.4722,"source_primary_group_share":0.5417,"capability_breadth":2.55,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5833,"capability_primary_group_count":11,"capability_non_primary_group_count":13},{"model":"suzume-llama-3-8B-multilingual","provider":"Lightblue","country":"JP","openness":"open","release_date":"2024-05-01","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":22,"benchmark_coverage_score":10.33,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":4.48,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.2778,"source_primary_group_share":0.3182,"capability_breadth":2.58,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":11,"capability_non_primary_group_count":11},{"model":"GPT-5.6 Sol (Non-reasoning)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":16,"benchmark_coverage_score":12.86,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.44,"source_primary":"artificialanalysis.ai","source_primary_share":0.9143,"source_primary_group_share":0.8125,"capability_breadth":3.22,"capability_primary":"code","capability_primary_share":0.4,"capability_primary_group_count":5,"capability_non_primary_group_count":11},{"model":"GPT-5.6 Terra (Non-reasoning)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-07-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":16,"benchmark_coverage_score":12.86,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.44,"source_primary":"artificialanalysis.ai","source_primary_share":0.9143,"source_primary_group_share":0.8125,"capability_breadth":3.22,"capability_primary":"code","capability_primary_share":0.4,"capability_primary_group_count":5,"capability_non_primary_group_count":11},{"model":"DeepSeek V4 Pro (Non-reasoning)","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2026-04-23","param_count":1600.0,"active_param_count":49.0,"training_compute_flop":9.702e+24,"num_benchmarks":36,"benchmark_group_count":18,"benchmark_coverage_score":12.1,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.57,"source_primary":"artificialanalysis.ai","source_primary_share":0.8857,"source_primary_group_share":0.7778,"capability_breadth":3.02,"capability_primary":"code","capability_primary_share":0.4286,"capability_primary_group_count":6,"capability_non_primary_group_count":12},{"model":"DeepSeek V4 Flash (Non-reasoning)","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2026-04-23","param_count":284.0,"active_param_count":13.0,"training_compute_flop":2.496e+24,"num_benchmarks":36,"benchmark_group_count":18,"benchmark_coverage_score":12.1,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.57,"source_primary":"artificialanalysis.ai","source_primary_share":0.8857,"source_primary_group_share":0.7778,"capability_breadth":3.02,"capability_primary":"code","capability_primary_share":0.4286,"capability_primary_group_count":6,"capability_non_primary_group_count":12},{"model":"Apriel-v1.6-15B-Thinker","provider":"ServiceNow","country":"CA","openness":"open","release_date":"2025-09-01","param_count":15.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":19,"benchmark_coverage_score":13.16,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.23,"source_primary":"artificialanalysis.ai","source_primary_share":0.9444,"source_primary_group_share":0.8947,"capability_breadth":3.29,"capability_primary":"code","capability_primary_share":0.4167,"capability_primary_group_count":6,"capability_non_primary_group_count":13},{"model":"Nemotron 3 Nano Omni 30B-A3B (Reasoning)","provider":"NVIDIA","country":"US","openness":"open","release_date":"2026-04-28","param_count":30.0,"active_param_count":3.0,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":16,"benchmark_coverage_score":13.94,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.29,"source_primary":"artificialanalysis.ai","source_primary_share":0.9444,"source_primary_group_share":0.875,"capability_breadth":3.48,"capability_primary":"code","capability_primary_share":0.4167,"capability_primary_group_count":6,"capability_non_primary_group_count":10},{"model":"NVIDIA Nemotron 3 Nano 4B","provider":"NVIDIA","country":"US","openness":"open","release_date":"2026-03-16","param_count":4.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":17,"benchmark_coverage_score":14.56,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.26,"source_primary":"artificialanalysis.ai","source_primary_share":0.9444,"source_primary_group_share":0.8824,"capability_breadth":3.64,"capability_primary":"code","capability_primary_share":0.3889,"capability_primary_group_count":5,"capability_non_primary_group_count":12},{"model":"Granite 4.1 3B","provider":"IBM","country":"US","openness":"open","release_date":"2026-04-29","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":17,"benchmark_coverage_score":14.56,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.26,"source_primary":"artificialanalysis.ai","source_primary_share":0.9444,"source_primary_group_share":0.8824,"capability_breadth":3.64,"capability_primary":"code","capability_primary_share":0.3889,"capability_primary_group_count":5,"capability_non_primary_group_count":12},{"model":"Hermes 4 405B (Non-reasoning)","provider":"Nous Research","country":"US","openness":"open","release_date":"2025-08-26","param_count":405.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":19,"benchmark_coverage_score":13.16,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.23,"source_primary":"artificialanalysis.ai","source_primary_share":0.9444,"source_primary_group_share":0.8947,"capability_breadth":3.29,"capability_primary":"code","capability_primary_share":0.4167,"capability_primary_group_count":6,"capability_non_primary_group_count":13},{"model":"Hermes 4 70B (Non-reasoning)","provider":"Nous Research","country":"US","openness":"open","release_date":"2025-08-26","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":19,"benchmark_coverage_score":13.16,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.23,"source_primary":"artificialanalysis.ai","source_primary_share":0.9444,"source_primary_group_share":0.8947,"capability_breadth":3.29,"capability_primary":"code","capability_primary_share":0.4167,"capability_primary_group_count":6,"capability_non_primary_group_count":13},{"model":"Exaone 4.0 1.2B (Reasoning)","provider":"LG AI","country":"KR","openness":"open","release_date":"2025-07-15","param_count":1.2,"active_param_count":null,"training_compute_flop":8.65e+22,"num_benchmarks":36,"benchmark_group_count":19,"benchmark_coverage_score":13.16,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.23,"source_primary":"artificialanalysis.ai","source_primary_share":0.9444,"source_primary_group_share":0.8947,"capability_breadth":3.29,"capability_primary":"code","capability_primary_share":0.4167,"capability_primary_group_count":6,"capability_non_primary_group_count":13},{"model":"Exaone 4.0 1.2B (Non-reasoning)","provider":"LG AI","country":"KR","openness":"open","release_date":"2025-07-15","param_count":1.2,"active_param_count":null,"training_compute_flop":8.65e+22,"num_benchmarks":36,"benchmark_group_count":19,"benchmark_coverage_score":13.16,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.23,"source_primary":"artificialanalysis.ai","source_primary_share":0.9444,"source_primary_group_share":0.8947,"capability_breadth":3.29,"capability_primary":"code","capability_primary_share":0.4167,"capability_primary_group_count":6,"capability_non_primary_group_count":13},{"model":"Motif-2-12.7B (Reasoning)","provider":"Motif","country":"US","openness":"open","release_date":"2025-11-07","param_count":12.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":19,"benchmark_coverage_score":13.16,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.23,"source_primary":"artificialanalysis.ai","source_primary_share":0.9444,"source_primary_group_share":0.8947,"capability_breadth":3.29,"capability_primary":"code","capability_primary_share":0.4167,"capability_primary_group_count":6,"capability_non_primary_group_count":13},{"model":"Jamba Reasoning 3B","provider":"AI21 Labs","country":"IL","openness":"open","release_date":"2025-10-08","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":19,"benchmark_coverage_score":13.16,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.23,"source_primary":"artificialanalysis.ai","source_primary_share":0.9444,"source_primary_group_share":0.8947,"capability_breadth":3.29,"capability_primary":"code","capability_primary_share":0.4167,"capability_primary_group_count":6,"capability_non_primary_group_count":13},{"model":"Qwen3.5 Omni Flash","provider":"Alibaba","country":"CN","openness":"api","release_date":"2026-03-30","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":18,"benchmark_coverage_score":15.07,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.57,"source_primary":"artificialanalysis.ai","source_primary_share":0.8889,"source_primary_group_share":0.7778,"capability_breadth":3.77,"capability_primary":"code","capability_primary_share":0.3889,"capability_primary_group_count":5,"capability_non_primary_group_count":13},{"model":"Ling-mini-2.0","provider":"InclusionAI","country":"CN","openness":"open","release_date":"2025-12-01","param_count":16.0,"active_param_count":1.4,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":19,"benchmark_coverage_score":13.16,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.23,"source_primary":"artificialanalysis.ai","source_primary_share":0.9444,"source_primary_group_share":0.8947,"capability_breadth":3.29,"capability_primary":"code","capability_primary_share":0.4167,"capability_primary_group_count":6,"capability_non_primary_group_count":13},{"model":"Llama 3.3 Nemotron Super 49B v1 (Non-reasoning)","provider":"NVIDIA","country":"US","openness":"open","release_date":"2025-03-18","param_count":49.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":19,"benchmark_coverage_score":12.96,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.23,"source_primary":"artificialanalysis.ai","source_primary_share":0.9444,"source_primary_group_share":0.8947,"capability_breadth":3.24,"capability_primary":"code","capability_primary_share":0.4167,"capability_primary_group_count":6,"capability_non_primary_group_count":13},{"model":"Sarvam M (Reasoning)","provider":"Sarvam AI","country":"IN","openness":"open","release_date":"2025-05-23","param_count":24.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":19,"benchmark_coverage_score":13.16,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.23,"source_primary":"artificialanalysis.ai","source_primary_share":0.9444,"source_primary_group_share":0.8947,"capability_breadth":3.29,"capability_primary":"code","capability_primary_share":0.4167,"capability_primary_group_count":6,"capability_non_primary_group_count":13},{"model":"InternVL3.5-8B","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2025-08-26","param_count":8.5,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":34,"benchmark_coverage_score":8.56,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.03,"source_primary":"benchmarklist.com","source_primary_share":0.4412,"source_primary_group_share":0.4412,"capability_breadth":2.14,"capability_primary":"vision_language","capability_primary_share":0.6471,"capability_primary_group_count":22,"capability_non_primary_group_count":12},{"model":"Seed 2.0 Pro","provider":"ByteDance","country":"CN","openness":"api","release_date":"2026-02-16","param_count":1000.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":33,"benchmark_coverage_score":23.14,"source_family_count":17,"source_group_family_count":15,"source_effective_breadth":4.24,"source_primary":"benchmarklist.com","source_primary_share":0.4167,"source_primary_group_share":0.4545,"capability_breadth":5.79,"capability_primary":"vision_language","capability_primary_share":0.2222,"capability_primary_group_count":8,"capability_non_primary_group_count":25},{"model":"DeepSeek V4 Flash (Max)","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2026-04-23","param_count":284.0,"active_param_count":13.0,"training_compute_flop":2.496e+24,"num_benchmarks":36,"benchmark_group_count":32,"benchmark_coverage_score":13.09,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":4.49,"source_primary":"matharena.ai","source_primary_share":0.3889,"source_primary_group_share":0.3438,"capability_breadth":3.27,"capability_primary":"math","capability_primary_share":0.5,"capability_primary_group_count":15,"capability_non_primary_group_count":17},{"model":"NeuralBeagle14-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-01-15","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":24,"benchmark_coverage_score":12.34,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"euroeval.com","source_primary_share":0.8333,"source_primary_group_share":0.75,"capability_breadth":3.09,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5,"capability_primary_group_count":14,"capability_non_primary_group_count":10},{"model":"Qwen 3 235B A22B 2507","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-07-01","param_count":235.0,"active_param_count":22.0,"training_compute_flop":4.752e+24,"num_benchmarks":36,"benchmark_group_count":29,"benchmark_coverage_score":17.82,"source_family_count":16,"source_group_family_count":16,"source_effective_breadth":8.17,"source_primary":"huggingface.co/spaces/gsma/open-telco-llm-leaderboard","source_primary_share":0.2286,"source_primary_group_share":0.2759,"capability_breadth":4.45,"capability_primary":"knowledge_general","capability_primary_share":0.3714,"capability_primary_group_count":11,"capability_non_primary_group_count":18},{"model":"Seed 2.1 Pro","provider":"ByteDance","country":"CN","openness":"api","release_date":"2026-06-24","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":36,"benchmark_group_count":34,"benchmark_coverage_score":18.25,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":1.46,"source_primary":"llm-stats.com","source_primary_share":0.8333,"source_primary_group_share":0.8235,"capability_breadth":4.56,"capability_primary":"vision_language","capability_primary_share":0.3611,"capability_primary_group_count":11,"capability_non_primary_group_count":23},{"model":"O3 (Medium)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-04-16","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":35,"benchmark_group_count":33,"benchmark_coverage_score":23.44,"source_family_count":17,"source_group_family_count":17,"source_effective_breadth":8.19,"source_primary":"hal.cs.princeton.edu","source_primary_share":0.3143,"source_primary_group_share":0.2727,"capability_breadth":5.86,"capability_primary":"agentic","capability_primary_share":0.2857,"capability_primary_group_count":8,"capability_non_primary_group_count":25},{"model":"GPT-3.5 Turbo (1106)","provider":"OpenAI","country":"US","openness":"api","release_date":"2023-11-06","param_count":175.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI GPT-3/GPT-3.5 family retired or superseded by newer API models.","num_benchmarks":35,"benchmark_group_count":31,"benchmark_coverage_score":21.67,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":6.54,"source_primary":"lk.instruct.kr","source_primary_share":0.2727,"source_primary_group_share":0.2258,"capability_breadth":5.42,"capability_primary":"instruction_following","capability_primary_share":0.2424,"capability_primary_group_count":8,"capability_non_primary_group_count":23},{"model":"Mistral Small (2402)","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-02-26","param_count":22.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":35,"benchmark_group_count":34,"benchmark_coverage_score":25.13,"source_family_count":8,"source_group_family_count":7,"source_effective_breadth":3.03,"source_primary":"livebench.ai","source_primary_share":0.4857,"source_primary_group_share":0.5,"capability_breadth":6.28,"capability_primary":"knowledge_general","capability_primary_share":0.2571,"capability_primary_group_count":8,"capability_non_primary_group_count":26},{"model":"Mixtral 8x7B","provider":"Mistral","country":"FR","openness":"open","release_date":"2023-12-11","param_count":46.7,"active_param_count":12.9,"training_compute_flop":7.74e+23,"num_benchmarks":35,"benchmark_group_count":28,"benchmark_coverage_score":13.73,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":4.9,"source_primary":"huggingface.co/spaces/speakleash/mt-bench-pl","source_primary_share":0.2571,"source_primary_group_share":0.3214,"capability_breadth":3.43,"capability_primary":"knowledge_general","capability_primary_share":0.4857,"capability_primary_group_count":11,"capability_non_primary_group_count":17},{"model":"Sailor2-8B-Chat","provider":"Sea AI Lab","country":"SG","openness":"open","release_date":"2024-12-03","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":35,"benchmark_group_count":25,"benchmark_coverage_score":9.98,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/meralion/seaeval_leaderboard","source_primary_share":0.9143,"source_primary_group_share":0.88,"capability_breadth":2.49,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5429,"capability_primary_group_count":13,"capability_non_primary_group_count":12},{"model":"GLM-5.2 (Non-reasoning)","provider":"Zhipu","country":"CN","openness":"open","release_date":"2026-06-13","param_count":744.0,"active_param_count":40.0,"training_compute_flop":null,"num_benchmarks":35,"benchmark_group_count":16,"benchmark_coverage_score":12.17,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.44,"source_primary":"artificialanalysis.ai","source_primary_share":0.9118,"source_primary_group_share":0.8125,"capability_breadth":3.04,"capability_primary":"code","capability_primary_share":0.4118,"capability_primary_group_count":5,"capability_non_primary_group_count":11},{"model":"Kimi K2.6 (Non-reasoning)","provider":"Moonshot","country":"CN","openness":"open","release_date":"2026-04-20","param_count":1000.0,"active_param_count":32.0,"training_compute_flop":null,"num_benchmarks":35,"benchmark_group_count":17,"benchmark_coverage_score":12.3,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.41,"source_primary":"artificialanalysis.ai","source_primary_share":0.9118,"source_primary_group_share":0.8235,"capability_breadth":3.07,"capability_primary":"code","capability_primary_share":0.4118,"capability_primary_group_count":5,"capability_non_primary_group_count":12},{"model":"MiMo-V2.5-Pro (Non-reasoning)","provider":"Xiaomi","country":"CN","openness":"open","release_date":"2026-04-23","param_count":1020.0,"active_param_count":42.0,"training_compute_flop":6.804e+24,"num_benchmarks":35,"benchmark_group_count":17,"benchmark_coverage_score":12.3,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.41,"source_primary":"artificialanalysis.ai","source_primary_share":0.9118,"source_primary_group_share":0.8235,"capability_breadth":3.07,"capability_primary":"code","capability_primary_share":0.4118,"capability_primary_group_count":5,"capability_non_primary_group_count":12},{"model":"EXAONE 4.5 33B","provider":"LG AI","country":"KR","openness":"open","release_date":"2026-04-09","param_count":33.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":35,"benchmark_group_count":17,"benchmark_coverage_score":14.45,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.41,"source_primary":"artificialanalysis.ai","source_primary_share":0.9143,"source_primary_group_share":0.8235,"capability_breadth":3.61,"capability_primary":"code","capability_primary_share":0.4,"capability_primary_group_count":5,"capability_non_primary_group_count":12},{"model":"GLM-5.1 (Non-reasoning)","provider":"Zhipu","country":"CN","openness":"open","release_date":"2026-03-27","param_count":744.0,"active_param_count":40.0,"training_compute_flop":null,"num_benchmarks":35,"benchmark_group_count":17,"benchmark_coverage_score":12.3,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.41,"source_primary":"artificialanalysis.ai","source_primary_share":0.9118,"source_primary_group_share":0.8235,"capability_breadth":3.07,"capability_primary":"code","capability_primary_share":0.4118,"capability_primary_group_count":5,"capability_non_primary_group_count":12},{"model":"GLM-4.7 Flash (Non-reasoning)","provider":"Zhipu","country":"CN","openness":"open","release_date":"2026-01-19","param_count":30.0,"active_param_count":3.0,"training_compute_flop":null,"num_benchmarks":35,"benchmark_group_count":17,"benchmark_coverage_score":12.3,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.41,"source_primary":"artificialanalysis.ai","source_primary_share":0.9118,"source_primary_group_share":0.8235,"capability_breadth":3.07,"capability_primary":"code","capability_primary_share":0.4118,"capability_primary_group_count":5,"capability_non_primary_group_count":12},{"model":"Claude Opus 4 (Thinking)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-05-22","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 4.0 family deprecated in favor of newer Claude 4.x models.","num_benchmarks":35,"benchmark_group_count":30,"benchmark_coverage_score":28.9,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":5.42,"source_primary":"artificialanalysis.ai","source_primary_share":0.3529,"source_primary_group_share":0.3667,"capability_breadth":7.22,"capability_primary":"vision_language","capability_primary_share":0.2059,"capability_primary_group_count":7,"capability_non_primary_group_count":23},{"model":"gpt-sw3-40B","provider":"AI Sweden","country":"SE","openness":"open","release_date":"2023-02-22","param_count":40.0,"active_param_count":null,"training_compute_flop":7.68e+22,"num_benchmarks":35,"benchmark_group_count":22,"benchmark_coverage_score":10.06,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.66,"source_primary":"euroeval.com","source_primary_share":0.8286,"source_primary_group_share":0.7273,"capability_breadth":2.52,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5714,"capability_primary_group_count":12,"capability_non_primary_group_count":10},{"model":"O4 Mini (2025-04-16) (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-04-16","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":34,"benchmark_group_count":30,"benchmark_coverage_score":20.69,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":5.84,"source_primary":"rank.opencompass.org.cn","source_primary_share":0.3125,"source_primary_group_share":0.3,"capability_breadth":5.17,"capability_primary":"code","capability_primary_share":0.3125,"capability_primary_group_count":9,"capability_non_primary_group_count":21},{"model":"davinci","provider":"OpenAI","country":"US","openness":"api","release_date":"2020-06-11","param_count":175.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI completions family retired or superseded by newer API models.","num_benchmarks":34,"benchmark_group_count":30,"benchmark_coverage_score":14.01,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.31,"source_primary":"crfm.stanford.edu","source_primary_share":0.8824,"source_primary_group_share":0.8667,"capability_breadth":3.5,"capability_primary":"knowledge_general","capability_primary_share":0.4412,"capability_primary_group_count":12,"capability_non_primary_group_count":18},{"model":"text-babbage-001","provider":"OpenAI","country":"US","openness":"api","release_date":"2022-01-27","param_count":1.3,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI completions family retired or superseded by newer API models.","num_benchmarks":34,"benchmark_group_count":30,"benchmark_coverage_score":14.01,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.31,"source_primary":"crfm.stanford.edu","source_primary_share":0.8824,"source_primary_group_share":0.8667,"capability_breadth":3.5,"capability_primary":"knowledge_general","capability_primary_share":0.4412,"capability_primary_group_count":12,"capability_non_primary_group_count":18},{"model":"text-ada-001","provider":"OpenAI","country":"US","openness":"api","release_date":"2022-01-27","param_count":0.35,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI completions family retired or superseded by newer API models.","num_benchmarks":34,"benchmark_group_count":30,"benchmark_coverage_score":14.01,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.31,"source_primary":"crfm.stanford.edu","source_primary_share":0.8824,"source_primary_group_share":0.8667,"capability_breadth":3.5,"capability_primary":"knowledge_general","capability_primary_share":0.4412,"capability_primary_group_count":12,"capability_non_primary_group_count":18},{"model":"Yi-1.5-6B","provider":"01.AI","country":"CN","openness":"open","release_date":"2024-05-11","param_count":6.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":34,"benchmark_group_count":25,"benchmark_coverage_score":13.44,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.34,"source_primary":"euroeval.com","source_primary_share":0.4412,"source_primary_group_share":0.4,"capability_breadth":3.36,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3824,"capability_primary_group_count":9,"capability_non_primary_group_count":16},{"model":"LFM2.5-8B-A1B","provider":"Liquid AI","country":"US","openness":"open","release_date":null,"param_count":8.0,"active_param_count":1.0,"training_compute_flop":null,"num_benchmarks":34,"benchmark_group_count":17,"benchmark_coverage_score":13.6,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.27,"source_primary":"artificialanalysis.ai","source_primary_share":0.9412,"source_primary_group_share":0.8824,"capability_breadth":3.4,"capability_primary":"code","capability_primary_share":0.4118,"capability_primary_group_count":5,"capability_non_primary_group_count":12},{"model":"Cogito v2.1 (Reasoning)","provider":"Nous Research","country":"US","openness":"open","release_date":"2025-09-01","param_count":671.0,"active_param_count":37.0,"training_compute_flop":null,"num_benchmarks":34,"benchmark_group_count":18,"benchmark_coverage_score":12.77,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.25,"source_primary":"artificialanalysis.ai","source_primary_share":0.9412,"source_primary_group_share":0.8889,"capability_breadth":3.19,"capability_primary":"code","capability_primary_share":0.4118,"capability_primary_group_count":6,"capability_non_primary_group_count":12},{"model":"GPT-5.5 Pro","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-04-23","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":34,"benchmark_group_count":29,"benchmark_coverage_score":15.36,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":6.05,"source_primary":"openai.com","source_primary_share":0.2759,"source_primary_group_share":0.2759,"capability_breadth":3.84,"capability_primary":"knowledge_general","capability_primary_share":0.4483,"capability_primary_group_count":13,"capability_non_primary_group_count":16},{"model":"GPT-5 (Thinking)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-08-07","param_count":null,"active_param_count":null,"training_compute_flop":6.6e+25,"num_benchmarks":34,"benchmark_group_count":29,"benchmark_coverage_score":13.68,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":1.86,"source_primary":"cdn.openai.com","source_primary_share":0.6774,"source_primary_group_share":0.7241,"capability_breadth":3.42,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4516,"capability_primary_group_count":14,"capability_non_primary_group_count":15},{"model":"Claude Opus 4.1 (20250805) (Thinking)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-08-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":34,"benchmark_group_count":30,"benchmark_coverage_score":20.08,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.26,"source_primary":"vals.ai","source_primary_share":0.4062,"source_primary_group_share":0.4333,"capability_breadth":5.02,"capability_primary":"knowledge_general","capability_primary_share":0.3438,"capability_primary_group_count":11,"capability_non_primary_group_count":19},{"model":"Gemini 3 Flash (High)","provider":"Google","country":"US","openness":"api","release_date":"2025-12-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":34,"benchmark_group_count":29,"benchmark_coverage_score":8.92,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":2.87,"source_primary":"ppbench.com","source_primary_share":0.5484,"source_primary_group_share":0.5517,"capability_breadth":2.23,"capability_primary":"reasoning_logical","capability_primary_share":0.6452,"capability_primary_group_count":18,"capability_non_primary_group_count":11},{"model":"Qwen3 Coder","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-07-23","param_count":235.0,"active_param_count":35.0,"training_compute_flop":null,"num_benchmarks":34,"benchmark_group_count":29,"benchmark_coverage_score":9.84,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":2.7,"source_primary":"ppbench.com","source_primary_share":0.5667,"source_primary_group_share":0.5517,"capability_breadth":2.46,"capability_primary":"reasoning_logical","capability_primary_share":0.5333,"capability_primary_group_count":15,"capability_non_primary_group_count":14},{"model":"Grok Beta","provider":"xAI","country":"US","openness":"api","release_date":"2024-11-04","param_count":270.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":33,"benchmark_group_count":26,"benchmark_coverage_score":16.95,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":5.83,"source_primary":"fab-ai.org","source_primary_share":0.2424,"source_primary_group_share":0.2692,"capability_breadth":4.24,"capability_primary":"knowledge_general","capability_primary_share":0.3939,"capability_primary_group_count":8,"capability_non_primary_group_count":18},{"model":"Grok 2","provider":"xAI","country":"US","openness":"api","release_date":"2024-08-13","param_count":270.0,"active_param_count":null,"training_compute_flop":2.96e+25,"num_benchmarks":33,"benchmark_group_count":26,"benchmark_coverage_score":23.29,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":6.9,"source_primary":"fab-ai.org","source_primary_share":0.2424,"source_primary_group_share":0.3077,"capability_breadth":5.82,"capability_primary":"knowledge_general","capability_primary_share":0.3333,"capability_primary_group_count":5,"capability_non_primary_group_count":21},{"model":"GPT-5 Pro","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-08-07","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":33,"benchmark_group_count":28,"benchmark_coverage_score":18.22,"source_family_count":16,"source_group_family_count":16,"source_effective_breadth":9.33,"source_primary":"scale.com","source_primary_share":0.2258,"source_primary_group_share":0.2143,"capability_breadth":4.55,"capability_primary":"reasoning_logical","capability_primary_share":0.3548,"capability_primary_group_count":9,"capability_non_primary_group_count":19},{"model":"GPT-3.5","provider":"OpenAI","country":"US","openness":"api","release_date":"2022-11-30","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI GPT-3/GPT-3.5 family retired or superseded by newer API models.","num_benchmarks":33,"benchmark_group_count":31,"benchmark_coverage_score":15.44,"source_family_count":14,"source_group_family_count":14,"source_effective_breadth":6.36,"source_primary":"huggingface.co/spaces/nexusflow/nexus_function_calling_leaderboard","source_primary_share":0.2903,"source_primary_group_share":0.2903,"capability_breadth":3.86,"capability_primary":"code","capability_primary_share":0.3871,"capability_primary_group_count":12,"capability_non_primary_group_count":19},{"model":"Yi-1.5-6B-Chat","provider":"01.AI","country":"CN","openness":"open","release_date":"2024-05-11","param_count":6.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":33,"benchmark_group_count":26,"benchmark_coverage_score":13.32,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":3.19,"source_primary":"huggingface.co","source_primary_share":0.5152,"source_primary_group_share":0.5,"capability_breadth":3.33,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4848,"capability_primary_group_count":10,"capability_non_primary_group_count":16},{"model":"Grok 3 Mini (High)","provider":"xAI","country":"US","openness":"api","release_date":"2025-02-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":33,"benchmark_group_count":30,"benchmark_coverage_score":16.0,"source_family_count":16,"source_group_family_count":16,"source_effective_breadth":6.52,"source_primary":"vals.ai","source_primary_share":0.3125,"source_primary_group_share":0.3333,"capability_breadth":4.0,"capability_primary":"math","capability_primary_share":0.375,"capability_primary_group_count":10,"capability_non_primary_group_count":20},{"model":"Sarvam 30B (High)","provider":"Sarvam AI","country":"IN","openness":"api","release_date":"2026-02-18","param_count":30.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":33,"benchmark_group_count":16,"benchmark_coverage_score":13.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"artificialanalysis.ai","source_primary_share":0.9394,"source_primary_group_share":0.875,"capability_breadth":3.25,"capability_primary":"code","capability_primary_share":0.4242,"capability_primary_group_count":5,"capability_non_primary_group_count":11},{"model":"Tri-21B-Think","provider":"Trillion Labs","country":"KR","openness":"open","release_date":"2026-02-10","param_count":21.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":33,"benchmark_group_count":16,"benchmark_coverage_score":13.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"artificialanalysis.ai","source_primary_share":0.9394,"source_primary_group_share":0.875,"capability_breadth":3.25,"capability_primary":"code","capability_primary_share":0.4242,"capability_primary_group_count":5,"capability_non_primary_group_count":11},{"model":"Sarvam 105B (High)","provider":"Sarvam AI","country":"IN","openness":"open","release_date":"2026-02-18","param_count":105.0,"active_param_count":9.0,"training_compute_flop":null,"num_benchmarks":33,"benchmark_group_count":16,"benchmark_coverage_score":13.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"artificialanalysis.ai","source_primary_share":0.9394,"source_primary_group_share":0.875,"capability_breadth":3.25,"capability_primary":"code","capability_primary_share":0.4242,"capability_primary_group_count":5,"capability_non_primary_group_count":11},{"model":"Tri-21B-Think Preview","provider":"Trillion Labs","country":"KR","openness":"open","release_date":"2026-02-10","param_count":21.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":33,"benchmark_group_count":16,"benchmark_coverage_score":13.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"artificialanalysis.ai","source_primary_share":0.9394,"source_primary_group_share":0.875,"capability_breadth":3.25,"capability_primary":"code","capability_primary_share":0.4242,"capability_primary_group_count":5,"capability_non_primary_group_count":11},{"model":"MiniCPM5-1B (Non-reasoning)","provider":"OpenBMB","country":"CN","openness":"open","release_date":"2026-05-21","param_count":1.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":33,"benchmark_group_count":16,"benchmark_coverage_score":13.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"artificialanalysis.ai","source_primary_share":0.9394,"source_primary_group_share":0.875,"capability_breadth":3.25,"capability_primary":"code","capability_primary_share":0.4242,"capability_primary_group_count":5,"capability_non_primary_group_count":11},{"model":"JT-35B-Flash","provider":"China Mobile","country":"CN","openness":"api","release_date":"2026-05-14","param_count":35.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":33,"benchmark_group_count":16,"benchmark_coverage_score":13.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"artificialanalysis.ai","source_primary_share":0.9394,"source_primary_group_share":0.875,"capability_breadth":3.25,"capability_primary":"code","capability_primary_share":0.4242,"capability_primary_group_count":5,"capability_non_primary_group_count":11},{"model":"Qwen 3 4B Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":4.0,"active_param_count":null,"training_compute_flop":8.64e+23,"num_benchmarks":33,"benchmark_group_count":23,"benchmark_coverage_score":10.11,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":3.09,"source_primary":"huggingface.co/spaces/yonyou-ai-lab/lom-benchmark-leaderboard","source_primary_share":0.5758,"source_primary_group_share":0.4783,"capability_breadth":2.53,"capability_primary":"reasoning_logical","capability_primary_share":0.6061,"capability_primary_group_count":15,"capability_non_primary_group_count":8},{"model":"O4 Mini (Medium)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-04-16","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":32,"benchmark_group_count":29,"benchmark_coverage_score":25.12,"source_family_count":16,"source_group_family_count":16,"source_effective_breadth":11.52,"source_primary":"matharena.ai","source_primary_share":0.1613,"source_primary_group_share":0.1379,"capability_breadth":6.28,"capability_primary":"math","capability_primary_share":0.2903,"capability_primary_group_count":7,"capability_non_primary_group_count":22},{"model":"Phi-3-small-128k-instruct","provider":"Microsoft","country":"US","openness":"open","release_date":"2024-05-07","param_count":7.4,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":32,"benchmark_group_count":28,"benchmark_coverage_score":28.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.38,"source_primary":"livebench.ai","source_primary_share":0.5312,"source_primary_group_share":0.6071,"capability_breadth":7.11,"capability_primary":"knowledge_multilingual","capability_primary_share":0.1875,"capability_primary_group_count":6,"capability_non_primary_group_count":22},{"model":"Mistral Medium","provider":"Mistral","country":"FR","openness":"api","release_date":"2023-12-11","param_count":123.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":32,"benchmark_group_count":28,"benchmark_coverage_score":21.33,"source_family_count":17,"source_group_family_count":16,"source_effective_breadth":8.34,"source_primary":"artificialanalysis.ai","source_primary_share":0.2188,"source_primary_group_share":0.25,"capability_breadth":5.33,"capability_primary":"knowledge_general","capability_primary_share":0.3438,"capability_primary_group_count":9,"capability_non_primary_group_count":19},{"model":"Nous-Hermes-2-Mistral-7B-DPO","provider":"Nous Research","country":"US","openness":"open","release_date":"2024-01-15","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":32,"benchmark_group_count":22,"benchmark_coverage_score":17.07,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.84,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.3125,"source_primary_group_share":0.3182,"capability_breadth":4.27,"capability_primary":"knowledge_general","capability_primary_share":0.375,"capability_primary_group_count":7,"capability_non_primary_group_count":15},{"model":"Grok 3 Mini Beta","provider":"xAI","country":"US","openness":"api","release_date":"2025-02-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":32,"benchmark_group_count":30,"benchmark_coverage_score":28.44,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":7.63,"source_primary":"storage.googleapis.com","source_primary_share":0.1875,"source_primary_group_share":0.2,"capability_breadth":7.11,"capability_primary":"instruction_following","capability_primary_share":0.25,"capability_primary_group_count":8,"capability_non_primary_group_count":22},{"model":"Claude Haiku 4.5 (20251001) (Thinking)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-10-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":32,"benchmark_group_count":29,"benchmark_coverage_score":16.08,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.43,"source_primary":"vals.ai","source_primary_share":0.8387,"source_primary_group_share":0.8276,"capability_breadth":4.02,"capability_primary":"knowledge_general","capability_primary_share":0.3871,"capability_primary_group_count":12,"capability_non_primary_group_count":17},{"model":"starcoder2-15B","provider":"BigCode","country":"FR","openness":"open","release_date":"2024-02-28","param_count":15.0,"active_param_count":null,"training_compute_flop":3.87e+23,"num_benchmarks":31,"benchmark_group_count":20,"benchmark_coverage_score":10.48,"source_family_count":10,"source_group_family_count":9,"source_effective_breadth":5.26,"source_primary":"evo-eval.github.io","source_primary_share":0.2759,"source_primary_group_share":0.3,"capability_breadth":2.62,"capability_primary":"code","capability_primary_share":0.5862,"capability_primary_group_count":8,"capability_non_primary_group_count":12},{"model":"mpt-7B","provider":"Databricks","country":"US","openness":"open","release_date":"2023-05-05","param_count":7.0,"active_param_count":null,"training_compute_flop":4.2e+22,"num_benchmarks":31,"benchmark_group_count":23,"benchmark_coverage_score":21.24,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.02,"source_primary":"epoch.ai","source_primary_share":0.3871,"source_primary_group_share":0.4783,"capability_breadth":5.31,"capability_primary":"knowledge_general","capability_primary_share":0.2903,"capability_primary_group_count":8,"capability_non_primary_group_count":15},{"model":"AlphaMonarch-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-01","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":31,"benchmark_group_count":21,"benchmark_coverage_score":9.68,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.83,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.3226,"source_primary_group_share":0.3333,"capability_breadth":2.42,"capability_primary":"knowledge_general","capability_primary_share":0.5806,"capability_primary_group_count":9,"capability_non_primary_group_count":12},{"model":"Daredevil-8B-abliterated","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-01","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":31,"benchmark_group_count":21,"benchmark_coverage_score":9.68,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.83,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.3226,"source_primary_group_share":0.3333,"capability_breadth":2.42,"capability_primary":"knowledge_general","capability_primary_share":0.5806,"capability_primary_group_count":9,"capability_non_primary_group_count":12},{"model":"NeuralLLaMa-3-8B-ORPO-v0.3","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-01","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":31,"benchmark_group_count":21,"benchmark_coverage_score":9.68,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.83,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.3226,"source_primary_group_share":0.3333,"capability_breadth":2.42,"capability_primary":"knowledge_general","capability_primary_share":0.5806,"capability_primary_group_count":9,"capability_non_primary_group_count":12},{"model":"GLM-5V Turbo","provider":"Zhipu","country":"CN","openness":"api","release_date":"2026-04-01","param_count":744.0,"active_param_count":40.0,"training_compute_flop":null,"num_benchmarks":31,"benchmark_group_count":28,"benchmark_coverage_score":18.75,"source_family_count":15,"source_group_family_count":15,"source_effective_breadth":8.91,"source_primary":"benchmarklist.com","source_primary_share":0.1935,"source_primary_group_share":0.2143,"capability_breadth":4.69,"capability_primary":"vision_language","capability_primary_share":0.3871,"capability_primary_group_count":11,"capability_non_primary_group_count":17},{"model":"O4 Mini (Low)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-04-16","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":30,"benchmark_group_count":24,"benchmark_coverage_score":20.91,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":5.14,"source_primary":"hal.cs.princeton.edu","source_primary_share":0.3929,"source_primary_group_share":0.375,"capability_breadth":5.23,"capability_primary":"agentic","capability_primary_share":0.2857,"capability_primary_group_count":6,"capability_non_primary_group_count":18},{"model":"DeepSeek Coder V2 Lite Instruct","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2024-06-17","param_count":16.0,"active_param_count":2.4,"training_compute_flop":null,"num_benchmarks":30,"benchmark_group_count":30,"benchmark_coverage_score":19.78,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":2.71,"source_primary":"livebench.ai","source_primary_share":0.5667,"source_primary_group_share":0.5667,"capability_breadth":4.95,"capability_primary":"code","capability_primary_share":0.3333,"capability_primary_group_count":10,"capability_non_primary_group_count":20},{"model":"Qwen 1.5 110B Chat","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-04-25","param_count":110.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":30,"benchmark_group_count":30,"benchmark_coverage_score":24.66,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":2.71,"source_primary":"livebench.ai","source_primary_share":0.5667,"source_primary_group_share":0.5667,"capability_breadth":6.16,"capability_primary":"reasoning_logical","capability_primary_share":0.2333,"capability_primary_group_count":7,"capability_non_primary_group_count":23},{"model":"Llama 2 13B Chat","provider":"Meta","country":"US","openness":"open","release_date":"2023-07-18","param_count":13.0,"active_param_count":null,"training_compute_flop":1.6e+23,"num_benchmarks":30,"benchmark_group_count":22,"benchmark_coverage_score":15.38,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":6.37,"source_primary":"lk.instruct.kr","source_primary_share":0.3,"source_primary_group_share":0.3182,"capability_breadth":3.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4,"capability_primary_group_count":8,"capability_non_primary_group_count":14},{"model":"NeuralLLaMa-3-8B-DT-v0.1","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-01","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":30,"benchmark_group_count":20,"benchmark_coverage_score":9.14,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":3.51,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.3333,"source_primary_group_share":0.35,"capability_breadth":2.28,"capability_primary":"knowledge_general","capability_primary_share":0.6,"capability_primary_group_count":9,"capability_non_primary_group_count":11},{"model":"stablelm-2-1.6B","provider":"Stability AI","country":"UK","openness":"open","release_date":"2024-01-19","param_count":1.6,"active_param_count":null,"training_compute_flop":1.92e+22,"num_benchmarks":30,"benchmark_group_count":18,"benchmark_coverage_score":10.65,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.31,"source_primary":"euroeval.com","source_primary_share":0.4667,"source_primary_group_share":0.5556,"capability_breadth":2.66,"capability_primary":"knowledge_general","capability_primary_share":0.5333,"capability_primary_group_count":7,"capability_non_primary_group_count":11},{"model":"Qwen 3.5 Plus (2026-04-20)","provider":"Alibaba","country":"CN","openness":"api","release_date":"2026-02-16","param_count":397.0,"active_param_count":17.0,"training_compute_flop":null,"num_benchmarks":30,"benchmark_group_count":29,"benchmark_coverage_score":21.43,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.34,"source_primary":"benchmarklist.com","source_primary_share":0.8621,"source_primary_group_share":0.8621,"capability_breadth":5.36,"capability_primary":"vision_language","capability_primary_share":0.3103,"capability_primary_group_count":9,"capability_non_primary_group_count":20},{"model":"Lucie-7B","provider":"OpenLLM-France","country":"FR","openness":"open","release_date":"2024-10-10","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":30,"benchmark_group_count":19,"benchmark_coverage_score":9.57,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.76,"source_primary":"euroeval.com","source_primary_share":0.8,"source_primary_group_share":0.6842,"capability_breadth":2.39,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6,"capability_primary_group_count":12,"capability_non_primary_group_count":7},{"model":"DBRX","provider":"Databricks","country":"US","openness":"open","release_date":"2024-03-27","param_count":132.0,"active_param_count":36.0,"training_compute_flop":2.6e+24,"num_benchmarks":29,"benchmark_group_count":21,"benchmark_coverage_score":19.01,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":5.31,"source_primary":"huggingface.co/spaces/allenai/urial-bench","source_primary_share":0.3793,"source_primary_group_share":0.2857,"capability_breadth":4.75,"capability_primary":"instruction_following","capability_primary_share":0.3103,"capability_primary_group_count":7,"capability_non_primary_group_count":14},{"model":"starcoder2-7B","provider":"BigCode","country":"FR","openness":"open","release_date":"2024-02-28","param_count":7.0,"active_param_count":null,"training_compute_flop":1.55e+23,"num_benchmarks":29,"benchmark_group_count":18,"benchmark_coverage_score":11.35,"source_family_count":8,"source_group_family_count":7,"source_effective_breadth":4.38,"source_primary":"evo-eval.github.io","source_primary_share":0.2963,"source_primary_group_share":0.3333,"capability_breadth":2.84,"capability_primary":"code","capability_primary_share":0.5556,"capability_primary_group_count":6,"capability_non_primary_group_count":12},{"model":"tulu-2-dpo-70B","provider":"Allen AI","country":"US","openness":"open","release_date":"2023-11-12","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":29,"benchmark_group_count":28,"benchmark_coverage_score":12.69,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":6.13,"source_primary":"allenai.github.io","source_primary_share":0.2414,"source_primary_group_share":0.25,"capability_breadth":3.17,"capability_primary":"instruction_following","capability_primary_share":0.4483,"capability_primary_group_count":13,"capability_non_primary_group_count":15},{"model":"Meta-Llama-3-8B-Instruct-abliterated-v3","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-01","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":29,"benchmark_group_count":19,"benchmark_coverage_score":8.25,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.65,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.3571,"source_primary_group_share":0.3684,"capability_breadth":2.06,"capability_primary":"knowledge_general","capability_primary_share":0.6786,"capability_primary_group_count":10,"capability_non_primary_group_count":9},{"model":"c4ai-command-a-03-2025","provider":"Cohere","country":"CA","openness":"open","release_date":"2025-03-13","param_count":111.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":29,"benchmark_group_count":24,"benchmark_coverage_score":20.1,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":4.57,"source_primary":"huggingface.co","source_primary_share":0.3214,"source_primary_group_share":0.375,"capability_breadth":5.03,"capability_primary":"knowledge_multilingual","capability_primary_share":0.2857,"capability_primary_group_count":8,"capability_non_primary_group_count":16},{"model":"Claude Fable 5 (Max)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-06-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":29,"benchmark_group_count":27,"benchmark_coverage_score":17.25,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":4.53,"source_primary":"epoch.ai","source_primary_share":0.3103,"source_primary_group_share":0.3333,"capability_breadth":4.31,"capability_primary":"math","capability_primary_share":0.3793,"capability_primary_group_count":10,"capability_non_primary_group_count":17},{"model":"Claude Opus 4.7 (Thinking)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-04-16","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":29,"benchmark_group_count":25,"benchmark_coverage_score":25.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":4.63,"source_primary":"arena.ai","source_primary_share":0.3214,"source_primary_group_share":0.32,"capability_breadth":6.43,"capability_primary":"knowledge_multilingual","capability_primary_share":0.25,"capability_primary_group_count":7,"capability_non_primary_group_count":18},{"model":"deepseek-coder-6.7B-instruct","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2023-10-29","param_count":6.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":28,"benchmark_group_count":17,"benchmark_coverage_score":8.5,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":4.07,"source_primary":"evo-eval.github.io","source_primary_share":0.3077,"source_primary_group_share":0.4118,"capability_breadth":2.13,"capability_primary":"code","capability_primary_share":0.6538,"capability_primary_group_count":9,"capability_non_primary_group_count":8},{"model":"Qwen-7B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2023-08-01","param_count":7.0,"active_param_count":null,"training_compute_flop":1.01e+23,"num_benchmarks":28,"benchmark_group_count":17,"benchmark_coverage_score":17.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":4.45,"source_primary":"epoch.ai","source_primary_share":0.2857,"source_primary_group_share":0.4118,"capability_breadth":5.85,"capability_primary":"code","capability_primary_share":0.2857,"capability_primary_group_count":5,"capability_non_primary_group_count":12},{"model":"DeepSeek Coder V2","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2024-06-17","param_count":236.0,"active_param_count":21.0,"training_compute_flop":null,"num_benchmarks":28,"benchmark_group_count":28,"benchmark_coverage_score":24.89,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":2.6,"source_primary":"livebench.ai","source_primary_share":0.6071,"source_primary_group_share":0.6071,"capability_breadth":6.22,"capability_primary":"reasoning_logical","capability_primary_share":0.2143,"capability_primary_group_count":6,"capability_non_primary_group_count":22},{"model":"dolphin-2.9.1-llama-3-8B","provider":"Cognitive Computations","country":"US","openness":"open","release_date":"2024-04-25","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":28,"benchmark_group_count":19,"benchmark_coverage_score":9.62,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.67,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.3571,"source_primary_group_share":0.4737,"capability_breadth":2.4,"capability_primary":"knowledge_general","capability_primary_share":0.5714,"capability_primary_group_count":8,"capability_non_primary_group_count":11},{"model":"Gemini 3 Flash (Preview) (High)","provider":"Google","country":"US","openness":"api","release_date":"2025-12-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":28,"benchmark_group_count":24,"benchmark_coverage_score":24.0,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":4.5,"source_primary":"benchmarklist.com","source_primary_share":0.4074,"source_primary_group_share":0.4167,"capability_breadth":8.01,"capability_primary":"code","capability_primary_share":0.1852,"capability_primary_group_count":5,"capability_non_primary_group_count":19},{"model":"GLM-5 FP8","provider":"Zhipu","country":"CN","openness":"open","release_date":"2026-02-12","param_count":744.0,"active_param_count":40.0,"training_compute_flop":6.84e+24,"num_benchmarks":28,"benchmark_group_count":14,"benchmark_coverage_score":12.85,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.18,"source_primary":"huggingface.co/spaces/hpai-bsc/turtle-leaderboard","source_primary_share":0.4286,"source_primary_group_share":0.6429,"capability_breadth":3.21,"capability_primary":"code","capability_primary_share":0.4643,"capability_primary_group_count":4,"capability_non_primary_group_count":10},{"model":"Claude Sonnet 4.6 (Thinking)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-02-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":28,"benchmark_group_count":24,"benchmark_coverage_score":9.09,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":2.17,"source_primary":"ppbench.com","source_primary_share":0.68,"source_primary_group_share":0.6667,"capability_breadth":2.27,"capability_primary":"reasoning_logical","capability_primary_share":0.64,"capability_primary_group_count":15,"capability_non_primary_group_count":9},{"model":"Viking-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-02-16","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":28,"benchmark_group_count":19,"benchmark_coverage_score":7.8,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.63,"source_primary":"euroeval.com","source_primary_share":0.75,"source_primary_group_share":0.7368,"capability_breadth":1.95,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6786,"capability_primary_group_count":14,"capability_non_primary_group_count":5},{"model":"Jamba 1.5 Large","provider":"AI21 Labs","country":"IL","openness":"api","release_date":"2024-08-22","param_count":398.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":27,"benchmark_group_count":26,"benchmark_coverage_score":12.52,"source_family_count":9,"source_group_family_count":8,"source_effective_breadth":5.28,"source_primary":"artificialanalysis.ai","source_primary_share":0.2593,"source_primary_group_share":0.2692,"capability_breadth":3.13,"capability_primary":"knowledge_general","capability_primary_share":0.5185,"capability_primary_group_count":13,"capability_non_primary_group_count":13},{"model":"internlm2-chat-20B","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2024-01-17","param_count":20.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":27,"benchmark_group_count":22,"benchmark_coverage_score":19.06,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":5.38,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.2963,"source_primary_group_share":0.3182,"capability_breadth":4.76,"capability_primary":"knowledge_multilingual","capability_primary_share":0.2963,"capability_primary_group_count":5,"capability_non_primary_group_count":17},{"model":"Gemini 2.0 Flash Lite (Preview 02-05)","provider":"Google","country":"US","openness":"api","release_date":"2025-02-25","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 2.0 family deprecated in favor of newer Gemini models.","num_benchmarks":27,"benchmark_group_count":26,"benchmark_coverage_score":14.51,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.38,"source_primary":"huggingface.co/spaces/maum-ai/koffvqa-leaderboard","source_primary_share":0.4074,"source_primary_group_share":0.4231,"capability_breadth":3.63,"capability_primary":"vision_language","capability_primary_share":0.4444,"capability_primary_group_count":11,"capability_non_primary_group_count":15},{"model":"Starling-LM-7B-alpha","provider":"UC Berkeley","country":"US","openness":"open","release_date":"2023-11-25","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":27,"benchmark_group_count":23,"benchmark_coverage_score":13.52,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":4.23,"source_primary":"huggingface.co/spaces/speakleash/mt-bench-pl","source_primary_share":0.3462,"source_primary_group_share":0.3913,"capability_breadth":3.38,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5,"capability_primary_group_count":10,"capability_non_primary_group_count":13},{"model":"ArmoRM-Llama3-8B-v0.1","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-23","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":27,"benchmark_group_count":26,"benchmark_coverage_score":19.06,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.25,"source_primary":"huggingface.co","source_primary_share":0.4074,"source_primary_group_share":0.4231,"capability_breadth":4.76,"capability_primary":"knowledge_general","capability_primary_share":0.2963,"capability_primary_group_count":7,"capability_non_primary_group_count":19},{"model":"Kimi K2 (0711)","provider":"Moonshot","country":"CN","openness":"open","release_date":"2025-07-11","param_count":1000.0,"active_param_count":32.0,"training_compute_flop":2.976e+24,"num_benchmarks":27,"benchmark_group_count":25,"benchmark_coverage_score":18.11,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":5.63,"source_primary":"raw.githubusercontent.com/llm2014/llm_benchmark","source_primary_share":0.3333,"source_primary_group_share":0.36,"capability_breadth":4.53,"capability_primary":"code","capability_primary_share":0.4074,"capability_primary_group_count":10,"capability_non_primary_group_count":15},{"model":"Claude Sonnet 4.5 (High)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-09-29","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":27,"benchmark_group_count":25,"benchmark_coverage_score":18.81,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.73,"source_primary":"huggingface.co/spaces/colab-potsdam/clem-leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.72,"capability_breadth":4.7,"capability_primary":"reasoning_logical","capability_primary_share":0.2963,"capability_primary_group_count":8,"capability_non_primary_group_count":17},{"model":"Llama 3.2 90B Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2024-09-25","param_count":90.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":26,"benchmark_group_count":24,"benchmark_coverage_score":15.36,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.27,"source_primary":"balrogai.com","source_primary_share":0.4231,"source_primary_group_share":0.4167,"capability_breadth":3.84,"capability_primary":"vision_language","capability_primary_share":0.4231,"capability_primary_group_count":9,"capability_non_primary_group_count":15},{"model":"Claude Opus 4.7 (xHigh)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-04-16","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":26,"benchmark_group_count":24,"benchmark_coverage_score":10.09,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":4.06,"source_primary":"matharena.ai","source_primary_share":0.4231,"source_primary_group_share":0.375,"capability_breadth":2.52,"capability_primary":"math","capability_primary_share":0.5769,"capability_primary_group_count":13,"capability_non_primary_group_count":11},{"model":"pythia-2.8B","provider":"EleutherAI","country":"US","openness":"open","release_date":"2023-04-03","param_count":2.8,"active_param_count":null,"training_compute_flop":5.038e+21,"num_benchmarks":26,"benchmark_group_count":15,"benchmark_coverage_score":10.82,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.53,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.3846,"source_primary_group_share":0.4667,"capability_breadth":2.7,"capability_primary":"knowledge_general","capability_primary_share":0.4615,"capability_primary_group_count":7,"capability_non_primary_group_count":8},{"model":"pythia-1.4B","provider":"EleutherAI","country":"US","openness":"open","release_date":"2023-04-03","param_count":1.4,"active_param_count":null,"training_compute_flop":2.519e+21,"num_benchmarks":26,"benchmark_group_count":15,"benchmark_coverage_score":10.82,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.53,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.3846,"source_primary_group_share":0.4667,"capability_breadth":2.7,"capability_primary":"knowledge_general","capability_primary_share":0.4615,"capability_primary_group_count":7,"capability_non_primary_group_count":8},{"model":"tulu-2-dpo-7B","provider":"Allen AI","country":"US","openness":"open","release_date":"2023-11-12","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":26,"benchmark_group_count":25,"benchmark_coverage_score":11.56,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":5.0,"source_primary":"allenai.github.io","source_primary_share":0.2692,"source_primary_group_share":0.28,"capability_breadth":2.89,"capability_primary":"instruction_following","capability_primary_share":0.4615,"capability_primary_group_count":12,"capability_non_primary_group_count":13},{"model":"Yi-1.5-9B-Chat-16K","provider":"01.AI","country":"CN","openness":"open","release_date":"2024-05-15","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":26,"benchmark_group_count":20,"benchmark_coverage_score":11.52,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.57,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.32,"source_primary_group_share":0.35,"capability_breadth":2.88,"capability_primary":"knowledge_general","capability_primary_share":0.48,"capability_primary_group_count":8,"capability_non_primary_group_count":12},{"model":"Llama-3-Smaug-8B","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-01","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":26,"benchmark_group_count":17,"benchmark_coverage_score":8.67,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.18,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.3846,"source_primary_group_share":0.4118,"capability_breadth":2.17,"capability_primary":"knowledge_general","capability_primary_share":0.6538,"capability_primary_group_count":8,"capability_non_primary_group_count":9},{"model":"Qwen 2.5 Coder 7B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-19","param_count":7.0,"active_param_count":null,"training_compute_flop":2.511e+23,"num_benchmarks":26,"benchmark_group_count":26,"benchmark_coverage_score":18.52,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":4.57,"source_primary":"huggingface.co/spaces/gsma/open-telco-llm-leaderboard","source_primary_share":0.3077,"source_primary_group_share":0.3077,"capability_breadth":4.63,"capability_primary":"knowledge_general","capability_primary_share":0.3462,"capability_primary_group_count":9,"capability_non_primary_group_count":17},{"model":"Qwen 3.5 Plus (Thinking)","provider":"Alibaba","country":"CN","openness":"api","release_date":"2026-02-16","param_count":700.0,"active_param_count":17.0,"training_compute_flop":null,"num_benchmarks":26,"benchmark_group_count":25,"benchmark_coverage_score":19.31,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.51,"source_primary":"vals.ai","source_primary_share":0.5769,"source_primary_group_share":0.6,"capability_breadth":4.83,"capability_primary":"reasoning_logical","capability_primary_share":0.3077,"capability_primary_group_count":7,"capability_non_primary_group_count":18},{"model":"MiniMax M1 40k","provider":"MiniMax","country":"CN","openness":"open","release_date":"2025-06-01","param_count":456.0,"active_param_count":45.9,"training_compute_flop":4.1861931e+24,"num_benchmarks":26,"benchmark_group_count":23,"benchmark_coverage_score":23.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":3.17,"source_primary":"artificialanalysis.ai","source_primary_share":0.4615,"source_primary_group_share":0.4783,"capability_breadth":6.15,"capability_primary":"reasoning_logical","capability_primary_share":0.2308,"capability_primary_group_count":5,"capability_non_primary_group_count":18},{"model":"Arcee Trinity Large (Thinking)","provider":"Arcee AI","country":"US","openness":"open","release_date":"2026-04-01","param_count":400.0,"active_param_count":13.0,"training_compute_flop":1.326e+24,"num_benchmarks":26,"benchmark_group_count":24,"benchmark_coverage_score":23.31,"source_family_count":17,"source_group_family_count":17,"source_effective_breadth":12.52,"source_primary":"designarena.ai","source_primary_share":0.2308,"source_primary_group_share":0.1667,"capability_breadth":5.83,"capability_primary":"code","capability_primary_share":0.3077,"capability_primary_group_count":7,"capability_non_primary_group_count":17},{"model":"Laguna M.1","provider":"Poolside","country":"US","openness":"open","release_date":"2026-04-28","param_count":225.0,"active_param_count":23.0,"training_compute_flop":null,"num_benchmarks":26,"benchmark_group_count":25,"benchmark_coverage_score":13.52,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":2.34,"source_primary":"vals.ai","source_primary_share":0.6538,"source_primary_group_share":0.64,"capability_breadth":3.38,"capability_primary":"code","capability_primary_share":0.4231,"capability_primary_group_count":10,"capability_non_primary_group_count":15},{"model":"WestLake-7B-v2","provider":"Other","country":"XX","openness":"open","release_date":"2024-01-22","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":26,"benchmark_group_count":17,"benchmark_coverage_score":11.66,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.84,"source_primary":"euroeval.com","source_primary_share":0.7692,"source_primary_group_share":0.6471,"capability_breadth":2.91,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5,"capability_primary_group_count":9,"capability_non_primary_group_count":8},{"model":"Gemini 3.6 Flash","provider":"Google","country":"US","openness":"api","release_date":null,"param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":26,"benchmark_group_count":26,"benchmark_coverage_score":24.58,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":7.04,"source_primary":"vals.ai","source_primary_share":0.2692,"source_primary_group_share":0.2692,"capability_breadth":6.15,"capability_primary":"knowledge_general","capability_primary_share":0.2692,"capability_primary_group_count":7,"capability_non_primary_group_count":19},{"model":"Claude Opus 4.6 (Non-reasoning)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-02-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":25,"benchmark_group_count":22,"benchmark_coverage_score":21.33,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":5.04,"source_primary":"scale.com","source_primary_share":0.375,"source_primary_group_share":0.3636,"capability_breadth":5.33,"capability_primary":"reasoning_logical","capability_primary_share":0.3333,"capability_primary_group_count":7,"capability_non_primary_group_count":15},{"model":"starcoder2-3B","provider":"BigCode","country":"FR","openness":"open","release_date":"2024-02-28","param_count":3.0,"active_param_count":null,"training_compute_flop":5.94e+22,"num_benchmarks":25,"benchmark_group_count":18,"benchmark_coverage_score":13.83,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":4.38,"source_primary":"evo-eval.github.io","source_primary_share":0.3478,"source_primary_group_share":0.3333,"capability_breadth":3.46,"capability_primary":"code","capability_primary_share":0.4783,"capability_primary_group_count":6,"capability_non_primary_group_count":12},{"model":"internlm2-chat-7B","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2024-01-10","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":25,"benchmark_group_count":19,"benchmark_coverage_score":17.19,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":4.15,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.3333,"source_primary_group_share":0.3684,"capability_breadth":4.3,"capability_primary":"knowledge_multilingual","capability_primary_share":0.2917,"capability_primary_group_count":5,"capability_non_primary_group_count":14},{"model":"PLLuM-12B-chat","provider":"PLLuM","country":"PL","openness":"open","release_date":"2025-02-24","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":25,"benchmark_group_count":21,"benchmark_coverage_score":9.09,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":3.59,"source_primary":"huggingface.co/spaces/speakleash/mt-bench-pl","source_primary_share":0.36,"source_primary_group_share":0.4286,"capability_breadth":2.27,"capability_primary":"knowledge_multilingual","capability_primary_share":0.64,"capability_primary_group_count":12,"capability_non_primary_group_count":9},{"model":"CarbonBeagle-11B-truthy","provider":"Other","country":"XX","openness":"open","release_date":"2024-02-01","param_count":11.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":25,"benchmark_group_count":16,"benchmark_coverage_score":8.14,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.84,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.4,"source_primary_group_share":0.4375,"capability_breadth":2.04,"capability_primary":"knowledge_general","capability_primary_share":0.68,"capability_primary_group_count":8,"capability_non_primary_group_count":8},{"model":"Llama-3-SauerkrautLM-8B-Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-01","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":25,"benchmark_group_count":16,"benchmark_coverage_score":8.14,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.84,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.4,"source_primary_group_share":0.4375,"capability_breadth":2.04,"capability_primary":"knowledge_general","capability_primary_share":0.68,"capability_primary_group_count":8,"capability_non_primary_group_count":8},{"model":"internlm2-7B","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2024-01-12","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":25,"benchmark_group_count":20,"benchmark_coverage_score":14.29,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.57,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.32,"source_primary_group_share":0.35,"capability_breadth":3.57,"capability_primary":"knowledge_general","capability_primary_share":0.36,"capability_primary_group_count":8,"capability_non_primary_group_count":12},{"model":"CarbonVillain-en-10.7B-v4","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-01","param_count":10.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":25,"benchmark_group_count":16,"benchmark_coverage_score":8.14,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.84,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.4,"source_primary_group_share":0.4375,"capability_breadth":2.04,"capability_primary":"knowledge_general","capability_primary_share":0.68,"capability_primary_group_count":8,"capability_non_primary_group_count":8},{"model":"Sakura-SOLAR-Instruct-CarbonVillain-en-10.7B-v2-slerp","provider":"Other","country":"XX","openness":"open","release_date":"2024-02-01","param_count":10.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":25,"benchmark_group_count":16,"benchmark_coverage_score":8.14,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.84,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.4,"source_primary_group_share":0.4375,"capability_breadth":2.04,"capability_primary":"knowledge_general","capability_primary_share":0.68,"capability_primary_group_count":8,"capability_non_primary_group_count":8},{"model":"StopCarbon-10.7B-v5","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-01","param_count":10.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":25,"benchmark_group_count":16,"benchmark_coverage_score":8.14,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.84,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.4,"source_primary_group_share":0.4375,"capability_breadth":2.04,"capability_primary":"knowledge_general","capability_primary_share":0.68,"capability_primary_group_count":8,"capability_non_primary_group_count":8},{"model":"Barcenas-Llama3-8B-ORPO","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-01","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":25,"benchmark_group_count":16,"benchmark_coverage_score":8.14,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.84,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.4,"source_primary_group_share":0.4375,"capability_breadth":2.04,"capability_primary":"knowledge_general","capability_primary_share":0.68,"capability_primary_group_count":8,"capability_non_primary_group_count":8},{"model":"BeagSake-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-01","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":25,"benchmark_group_count":16,"benchmark_coverage_score":8.14,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.84,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.4,"source_primary_group_share":0.4375,"capability_breadth":2.04,"capability_primary":"knowledge_general","capability_primary_share":0.68,"capability_primary_group_count":8,"capability_non_primary_group_count":8},{"model":"Llama-3-8B-Instruct-abliterated-v2","provider":"Meta","country":"US","openness":"open","release_date":"2024-06-01","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":25,"benchmark_group_count":15,"benchmark_coverage_score":8.42,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.85,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.4,"source_primary_group_share":0.4667,"capability_breadth":2.1,"capability_primary":"knowledge_general","capability_primary_share":0.64,"capability_primary_group_count":7,"capability_non_primary_group_count":8},{"model":"LLaMAntino-3-ANITA-8B-Inst-DPO-ITA","provider":"Meta","country":"US","openness":"open","release_date":"2024-04-29","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":25,"benchmark_group_count":20,"benchmark_coverage_score":13.97,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.47,"source_primary":"huggingface.co","source_primary_share":0.44,"source_primary_group_share":0.55,"capability_breadth":3.49,"capability_primary":"knowledge_multilingual","capability_primary_share":0.44,"capability_primary_group_count":10,"capability_non_primary_group_count":10},{"model":"Ministral 8B","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-10-16","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":25,"benchmark_group_count":16,"benchmark_coverage_score":9.93,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":4.74,"source_primary":"fab-ai.org","source_primary_share":0.3478,"source_primary_group_share":0.375,"capability_breadth":2.48,"capability_primary":"knowledge_general","capability_primary_share":0.6087,"capability_primary_group_count":8,"capability_non_primary_group_count":8},{"model":"LongCat-Flash-Chat","provider":"Meituan","country":"CN","openness":"api","release_date":"2025-09-01","param_count":560.0,"active_param_count":27.0,"training_compute_flop":null,"num_benchmarks":25,"benchmark_group_count":24,"benchmark_coverage_score":23.36,"source_family_count":14,"source_group_family_count":14,"source_effective_breadth":8.47,"source_primary":"rank.opencompass.org.cn","source_primary_share":0.28,"source_primary_group_share":0.25,"capability_breadth":5.84,"capability_primary":"knowledge_general","capability_primary_share":0.32,"capability_primary_group_count":8,"capability_non_primary_group_count":16},{"model":"Grok 3 Mini (Low)","provider":"xAI","country":"US","openness":"api","release_date":"2025-02-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":25,"benchmark_group_count":23,"benchmark_coverage_score":14.97,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":4.17,"source_primary":"vals.ai","source_primary_share":0.4,"source_primary_group_share":0.4348,"capability_breadth":3.74,"capability_primary":"math","capability_primary_share":0.4,"capability_primary_group_count":8,"capability_non_primary_group_count":15},{"model":"granite-3.0-2B-base","provider":"IBM","country":"US","openness":"open","release_date":"2024-10-02","param_count":2.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":25,"benchmark_group_count":17,"benchmark_coverage_score":11.85,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.11,"source_primary":"euroeval.com","source_primary_share":0.6,"source_primary_group_share":0.5882,"capability_breadth":2.96,"capability_primary":"knowledge_multilingual","capability_primary_share":0.48,"capability_primary_group_count":7,"capability_non_primary_group_count":10},{"model":"stablelm-zephyr-3B","provider":"Stability AI","country":"UK","openness":"open","release_date":"2023-11-21","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":25,"benchmark_group_count":24,"benchmark_coverage_score":14.97,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.67,"source_primary":"euroeval.com","source_primary_share":0.52,"source_primary_group_share":0.5,"capability_breadth":3.74,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4,"capability_primary_group_count":10,"capability_non_primary_group_count":14},{"model":"Qwen2.5-Math-72B-Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-16","param_count":72.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":24,"benchmark_group_count":18,"benchmark_coverage_score":10.47,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.95,"source_primary":"huggingface.co/spaces/toloka/u-math-leaderboard","source_primary_share":0.3333,"source_primary_group_share":0.3333,"capability_breadth":2.62,"capability_primary":"math","capability_primary_share":0.5833,"capability_primary_group_count":10,"capability_non_primary_group_count":8},{"model":"Jamba 1.5 Mini","provider":"AI21 Labs","country":"IL","openness":"api","release_date":"2024-08-22","param_count":52.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":24,"benchmark_group_count":23,"benchmark_coverage_score":11.64,"source_family_count":10,"source_group_family_count":9,"source_effective_breadth":5.04,"source_primary":"artificialanalysis.ai","source_primary_share":0.2917,"source_primary_group_share":0.3043,"capability_breadth":2.91,"capability_primary":"knowledge_general","capability_primary_share":0.5417,"capability_primary_group_count":12,"capability_non_primary_group_count":11},{"model":"Hermes-2-Pro-Mistral-7B","provider":"Nous Research","country":"US","openness":"open","release_date":"2024-03-14","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":24,"benchmark_group_count":15,"benchmark_coverage_score":10.38,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.36,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.4167,"source_primary_group_share":0.4,"capability_breadth":2.59,"capability_primary":"knowledge_general","capability_primary_share":0.5417,"capability_primary_group_count":6,"capability_non_primary_group_count":9},{"model":"Nous-Hermes-2-Mixtral-8x7B-DPO","provider":"Nous Research","country":"US","openness":"open","release_date":"2024-01-11","param_count":56.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":24,"benchmark_group_count":23,"benchmark_coverage_score":17.72,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":4.23,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.3333,"source_primary_group_share":0.3043,"capability_breadth":4.43,"capability_primary":"knowledge_general","capability_primary_share":0.3333,"capability_primary_group_count":7,"capability_non_primary_group_count":16},{"model":"ChimeraLlama-3-8B-v3","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-01","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":24,"benchmark_group_count":15,"benchmark_coverage_score":7.58,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.53,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.4167,"source_primary_group_share":0.4667,"capability_breadth":1.89,"capability_primary":"knowledge_general","capability_primary_share":0.7083,"capability_primary_group_count":8,"capability_non_primary_group_count":7},{"model":"L-MChat-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-01","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":24,"benchmark_group_count":15,"benchmark_coverage_score":7.58,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.53,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.4167,"source_primary_group_share":0.4667,"capability_breadth":1.89,"capability_primary":"knowledge_general","capability_primary_share":0.7083,"capability_primary_group_count":8,"capability_non_primary_group_count":7},{"model":"Lexora-Medium-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-24","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":24,"benchmark_group_count":19,"benchmark_coverage_score":14.58,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.24,"source_primary":"huggingface.co","source_primary_share":0.4583,"source_primary_group_share":0.5789,"capability_breadth":3.65,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4167,"capability_primary_group_count":9,"capability_non_primary_group_count":10},{"model":"O1 Pro","provider":"OpenAI","country":"US","openness":"api","release_date":"2024-12-05","param_count":200.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":24,"benchmark_group_count":16,"benchmark_coverage_score":15.07,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":4.27,"source_primary":"scale.com","source_primary_share":0.3333,"source_primary_group_share":0.375,"capability_breadth":3.77,"capability_primary":"vision_language","capability_primary_share":0.3889,"capability_primary_group_count":7,"capability_non_primary_group_count":9},{"model":"Llama-3.1-SuperNova-Lite","provider":"Meta","country":"US","openness":"open","release_date":"2024-09-10","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":24,"benchmark_group_count":23,"benchmark_coverage_score":14.77,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.89,"source_primary":"huggingface.co","source_primary_share":0.4583,"source_primary_group_share":0.4783,"capability_breadth":3.69,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3333,"capability_primary_group_count":8,"capability_non_primary_group_count":15},{"model":"Hy3-preview","provider":"Tencent","country":"CN","openness":"open","release_date":"2026-04-24","param_count":295.0,"active_param_count":21.0,"training_compute_flop":null,"num_benchmarks":24,"benchmark_group_count":23,"benchmark_coverage_score":23.0,"source_family_count":15,"source_group_family_count":15,"source_effective_breadth":8.97,"source_primary":"benchmarklist.com","source_primary_share":0.25,"source_primary_group_share":0.2609,"capability_breadth":5.76,"capability_primary":"reasoning_logical","capability_primary_share":0.2917,"capability_primary_group_count":6,"capability_non_primary_group_count":17},{"model":"Nemotron-Mini-4B-Instruct","provider":"NVIDIA","country":"US","openness":"open","release_date":"2024-09-10","param_count":4.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":24,"benchmark_group_count":20,"benchmark_coverage_score":13.09,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.94,"source_primary":"huggingface.co","source_primary_share":0.7083,"source_primary_group_share":0.65,"capability_breadth":3.27,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4583,"capability_primary_group_count":8,"capability_non_primary_group_count":12},{"model":"Hy3-preview (High)","provider":"Tencent","country":"CN","openness":"open","release_date":"2026-04-24","param_count":295.0,"active_param_count":21.0,"training_compute_flop":null,"num_benchmarks":24,"benchmark_group_count":17,"benchmark_coverage_score":17.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.45,"source_primary":"rank.opencompass.org.cn","source_primary_share":0.875,"source_primary_group_share":0.8235,"capability_breadth":6.13,"capability_primary":"knowledge_general","capability_primary_share":0.2917,"capability_primary_group_count":6,"capability_non_primary_group_count":11},{"model":"Claude Opus 4.1 (Thinking)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-08-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":23,"benchmark_group_count":21,"benchmark_coverage_score":21.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":3.27,"source_primary":"artificialanalysis.ai","source_primary_share":0.5652,"source_primary_group_share":0.5238,"capability_breadth":6.08,"capability_primary":"vision_language","capability_primary_share":0.2609,"capability_primary_group_count":5,"capability_non_primary_group_count":16},{"model":"Qwen 2.5 Max","provider":"Alibaba","country":"CN","openness":"api","release_date":"2025-01-29","param_count":236.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":23,"benchmark_group_count":22,"benchmark_coverage_score":21.04,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":6.54,"source_primary":"artificialanalysis.ai","source_primary_share":0.3182,"source_primary_group_share":0.3182,"capability_breadth":5.26,"capability_primary":"knowledge_general","capability_primary_share":0.2727,"capability_primary_group_count":6,"capability_non_primary_group_count":16},{"model":"deepseek-coder-1.3B-instruct","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2023-10-29","param_count":1.3,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":23,"benchmark_group_count":15,"benchmark_coverage_score":10.2,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.57,"source_primary":"evo-eval.github.io","source_primary_share":0.381,"source_primary_group_share":0.4667,"capability_breadth":2.55,"capability_primary":"code","capability_primary_share":0.5714,"capability_primary_group_count":7,"capability_non_primary_group_count":8},{"model":"Gemini 2.0 Flash (01-21) (Thinking)","provider":"Google","country":"US","openness":"api","release_date":"2025-01-21","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 2.0 family deprecated in favor of newer Gemini models.","num_benchmarks":23,"benchmark_group_count":18,"benchmark_coverage_score":14.67,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":9.53,"source_primary":"huggingface.co/spaces/toloka/u-math-leaderboard","source_primary_share":0.3636,"source_primary_group_share":0.2222,"capability_breadth":3.67,"capability_primary":"math","capability_primary_share":0.4545,"capability_primary_group_count":6,"capability_non_primary_group_count":12},{"model":"Bielik-11B-v2.2-Instruct","provider":"SpeakLeash","country":"PL","openness":"open","release_date":"2024-08-26","param_count":11.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":23,"benchmark_group_count":20,"benchmark_coverage_score":9.84,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.28,"source_primary":"huggingface.co/spaces/speakleash/mt-bench-pl","source_primary_share":0.3913,"source_primary_group_share":0.45,"capability_breadth":2.46,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6087,"capability_primary_group_count":11,"capability_non_primary_group_count":9},{"model":"Bielik-11B-v2.1-Instruct","provider":"SpeakLeash","country":"PL","openness":"open","release_date":"2024-08-26","param_count":11.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":23,"benchmark_group_count":20,"benchmark_coverage_score":9.84,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.28,"source_primary":"huggingface.co/spaces/speakleash/mt-bench-pl","source_primary_share":0.3913,"source_primary_group_share":0.45,"capability_breadth":2.46,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6087,"capability_primary_group_count":11,"capability_non_primary_group_count":9},{"model":"glm-4-9B","provider":"Zhipu","country":"CN","openness":"open","release_date":"2024-06-04","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":23,"benchmark_group_count":14,"benchmark_coverage_score":13.44,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":3.5,"source_primary":"huggingface.co/spaces/yangzhang33/greekmmlu-leaderboard","source_primary_share":0.2727,"source_primary_group_share":0.4286,"capability_breadth":3.36,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4545,"capability_primary_group_count":5,"capability_non_primary_group_count":9},{"model":"NeuralDaredevil-8B-abliterated","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-27","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":23,"benchmark_group_count":18,"benchmark_coverage_score":14.67,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.38,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.3636,"source_primary_group_share":0.3889,"capability_breadth":3.67,"capability_primary":"knowledge_general","capability_primary_share":0.4091,"capability_primary_group_count":8,"capability_non_primary_group_count":10},{"model":"GPT-5.1 Codex Max","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-11-19","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":23,"benchmark_group_count":16,"benchmark_coverage_score":14.45,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":8.0,"source_primary":"sreskillsbench.com","source_primary_share":0.3182,"source_primary_group_share":0.25,"capability_breadth":3.61,"capability_primary":"code","capability_primary_share":0.3636,"capability_primary_group_count":7,"capability_non_primary_group_count":9},{"model":"Gemma 4 26B A4B (IT) (Reasoning)","provider":"Google","country":"US","openness":"open","release_date":"2026-04-02","param_count":25.2,"active_param_count":3.8,"training_compute_flop":null,"num_benchmarks":23,"benchmark_group_count":9,"benchmark_coverage_score":7.41,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.25,"source_primary":"huggingface.co/spaces/lang-uk/ukrainian-llm-leaderboard","source_primary_share":0.9,"source_primary_group_share":0.8889,"capability_breadth":1.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7,"capability_primary_group_count":6,"capability_non_primary_group_count":3},{"model":"openchat-3.6-8B-20240522 (2024-05-22)","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-07","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":22,"benchmark_group_count":18,"benchmark_coverage_score":15.12,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.6,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.3636,"source_primary_group_share":0.3889,"capability_breadth":3.78,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3636,"capability_primary_group_count":6,"capability_non_primary_group_count":12},{"model":"Bielik-11B-v2.0-Instruct","provider":"SpeakLeash","country":"PL","openness":"open","release_date":"2024-08-26","param_count":11.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":22,"benchmark_group_count":19,"benchmark_coverage_score":10.3,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.98,"source_primary":"huggingface.co/spaces/speakleash/mt-bench-pl","source_primary_share":0.4091,"source_primary_group_share":0.4737,"capability_breadth":2.57,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5909,"capability_primary_group_count":10,"capability_non_primary_group_count":9},{"model":"Gemini 1.5 Pro (Sept)","provider":"Google","country":"US","openness":"api","release_date":"2024-09-24","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 1.x family retired or superseded by newer Gemini models.","num_benchmarks":22,"benchmark_group_count":20,"benchmark_coverage_score":18.19,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":4.65,"source_primary":"artificialanalysis.ai","source_primary_share":0.381,"source_primary_group_share":0.35,"capability_breadth":4.55,"capability_primary":"reasoning_logical","capability_primary_share":0.381,"capability_primary_group_count":8,"capability_non_primary_group_count":12},{"model":"command-r7B-12-2024","provider":"Cohere","country":"CA","openness":"open","release_date":"2024-12-11","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":22,"benchmark_group_count":21,"benchmark_coverage_score":19.82,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.97,"source_primary":"huggingface.co/spaces/gsma/open-telco-llm-leaderboard","source_primary_share":0.381,"source_primary_group_share":0.381,"capability_breadth":4.96,"capability_primary":"knowledge_general","capability_primary_share":0.381,"capability_primary_group_count":8,"capability_non_primary_group_count":13},{"model":"OLMo 2 7B","provider":"Allen AI","country":"US","openness":"open","release_date":"2024-11-01","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":22,"benchmark_group_count":14,"benchmark_coverage_score":14.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.78,"source_primary":"artificialanalysis.ai","source_primary_share":0.5,"source_primary_group_share":0.7143,"capability_breadth":3.84,"capability_primary":"knowledge_general","capability_primary_share":0.4545,"capability_primary_group_count":4,"capability_non_primary_group_count":10},{"model":"Kimi Linear 48B A3B Instruct","provider":"Moonshot","country":"CN","openness":"open","release_date":"2025-10-30","param_count":48.0,"active_param_count":3.0,"training_compute_flop":null,"num_benchmarks":22,"benchmark_group_count":20,"benchmark_coverage_score":20.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.33,"source_primary":"artificialanalysis.ai","source_primary_share":0.5238,"source_primary_group_share":0.5,"capability_breadth":7.0,"capability_primary":"knowledge_general","capability_primary_share":0.2381,"capability_primary_group_count":5,"capability_non_primary_group_count":15},{"model":"Claude Sonnet 4 (20250514) (Thinking)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-05-14","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 4.0 family deprecated in favor of newer Claude 4.x models.","num_benchmarks":22,"benchmark_group_count":21,"benchmark_coverage_score":11.84,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.65,"source_primary":"vals.ai","source_primary_share":0.7619,"source_primary_group_share":0.7619,"capability_breadth":2.96,"capability_primary":"knowledge_general","capability_primary_share":0.5238,"capability_primary_group_count":11,"capability_non_primary_group_count":10},{"model":"Laguna XS.2","provider":"Poolside","country":"US","openness":"open","release_date":"2026-04-28","param_count":33.0,"active_param_count":3.0,"training_compute_flop":null,"num_benchmarks":22,"benchmark_group_count":21,"benchmark_coverage_score":12.74,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":1.69,"source_primary":"vals.ai","source_primary_share":0.7727,"source_primary_group_share":0.7619,"capability_breadth":3.18,"capability_primary":"code","capability_primary_share":0.4091,"capability_primary_group_count":8,"capability_non_primary_group_count":13},{"model":"Mistral-Nemo-Base-2407","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-07-18","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":22,"benchmark_group_count":14,"benchmark_coverage_score":13.08,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":3.5,"source_primary":"huggingface.co/spaces/yangzhang33/greekmmlu-leaderboard","source_primary_share":0.2727,"source_primary_group_share":0.4286,"capability_breadth":3.27,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5,"capability_primary_group_count":4,"capability_non_primary_group_count":10},{"model":"GPT-5.1 (Low)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-11-12","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":21,"benchmark_group_count":17,"benchmark_coverage_score":17.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":6.72,"source_primary":"designarena.ai","source_primary_share":0.3158,"source_primary_group_share":0.2353,"capability_breadth":7.37,"capability_primary":"vision_language","capability_primary_share":0.1579,"capability_primary_group_count":3,"capability_non_primary_group_count":14},{"model":"GPT-4 Turbo (Preview)","provider":"OpenAI","country":"US","openness":"api","release_date":"2023-11-06","param_count":1760.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI GPT-4/GPT-4 Turbo family retired or superseded by newer API models.","num_benchmarks":21,"benchmark_group_count":18,"benchmark_coverage_score":18.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":4.05,"source_primary":"scale.com","source_primary_share":0.4211,"source_primary_group_share":0.3889,"capability_breadth":6.56,"capability_primary":"knowledge_general","capability_primary_share":0.2105,"capability_primary_group_count":4,"capability_non_primary_group_count":14},{"model":"Gemini 1.5 Flash (0514)","provider":"Google","country":"US","openness":"api","release_date":"2024-05-15","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 1.x family retired or superseded by newer Gemini models.","num_benchmarks":21,"benchmark_group_count":21,"benchmark_coverage_score":21.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.49,"source_primary":"livebench.ai","source_primary_share":0.8095,"source_primary_group_share":0.8095,"capability_breadth":6.58,"capability_primary":"math","capability_primary_share":0.1905,"capability_primary_group_count":4,"capability_non_primary_group_count":17},{"model":"Qwen 2.5 Coder 32B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-11-12","param_count":32.0,"active_param_count":null,"training_compute_flop":1.073e+24,"num_benchmarks":21,"benchmark_group_count":21,"benchmark_coverage_score":17.82,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":4.28,"source_primary":"epoch.ai","source_primary_share":0.2857,"source_primary_group_share":0.2857,"capability_breadth":4.45,"capability_primary":"code","capability_primary_share":0.3333,"capability_primary_group_count":7,"capability_non_primary_group_count":14},{"model":"Mixtral-8x22B-v0.1","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-04-10","param_count":176.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":21,"benchmark_group_count":14,"benchmark_coverage_score":13.89,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":3.77,"source_primary":"huggingface.co/spaces/aim-harvard/rabbits-leaderboard","source_primary_share":0.3333,"source_primary_group_share":0.4286,"capability_breadth":3.47,"capability_primary":"knowledge_general","capability_primary_share":0.4762,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Qwen 1.5 72B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-02-05","param_count":72.0,"active_param_count":null,"training_compute_flop":1.3e+24,"num_benchmarks":21,"benchmark_group_count":16,"benchmark_coverage_score":11.53,"source_family_count":7,"source_group_family_count":6,"source_effective_breadth":3.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.381,"source_primary_group_share":0.4375,"capability_breadth":2.88,"capability_primary":"knowledge_general","capability_primary_share":0.4762,"capability_primary_group_count":8,"capability_non_primary_group_count":8},{"model":"Pixtral-12B-2409","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-09-11","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":21,"benchmark_group_count":17,"benchmark_coverage_score":8.05,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.14,"source_primary":"huggingface.co/spaces/maum-ai/koffvqa-leaderboard","source_primary_share":0.5238,"source_primary_group_share":0.6471,"capability_breadth":2.01,"capability_primary":"vision_language","capability_primary_share":0.619,"capability_primary_group_count":13,"capability_non_primary_group_count":4},{"model":"Gemini 1.5 Pro (Preview 0801)","provider":"Google","country":"US","openness":"api","release_date":"2024-08-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 1.x family retired or superseded by newer Gemini models.","num_benchmarks":21,"benchmark_group_count":21,"benchmark_coverage_score":21.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":1.51,"source_primary":"livebench.ai","source_primary_share":0.8095,"source_primary_group_share":0.8095,"capability_breadth":5.73,"capability_primary":"structured_precision","capability_primary_share":0.2381,"capability_primary_group_count":5,"capability_non_primary_group_count":16},{"model":"PLLuM-12B-nc-chat","provider":"PLLuM","country":"PL","openness":"open","release_date":"2025-02-24","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":21,"benchmark_group_count":18,"benchmark_coverage_score":10.82,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.7,"source_primary":"huggingface.co/spaces/speakleash/mt-bench-pl","source_primary_share":0.4286,"source_primary_group_share":0.5,"capability_breadth":2.71,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5714,"capability_primary_group_count":9,"capability_non_primary_group_count":9},{"model":"Configurable-Yi-1.5-9B-Chat","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-12","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":21,"benchmark_group_count":19,"benchmark_coverage_score":12.88,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":3.25,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.381,"source_primary_group_share":0.3684,"capability_breadth":3.22,"capability_primary":"knowledge_general","capability_primary_share":0.381,"capability_primary_group_count":7,"capability_non_primary_group_count":12},{"model":"Faro-Yi-9B-DPO","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-07","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":21,"benchmark_group_count":19,"benchmark_coverage_score":12.34,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":3.25,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.381,"source_primary_group_share":0.3684,"capability_breadth":3.08,"capability_primary":"reasoning_logical","capability_primary_share":0.4762,"capability_primary_group_count":9,"capability_non_primary_group_count":10},{"model":"Hermes-2-Theta-Llama-3-8B","provider":"Nous Research","country":"US","openness":"open","release_date":"2024-05-05","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":21,"benchmark_group_count":17,"benchmark_coverage_score":15.34,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":3.25,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.381,"source_primary_group_share":0.4118,"capability_breadth":3.83,"capability_primary":"knowledge_general","capability_primary_share":0.3333,"capability_primary_group_count":6,"capability_non_primary_group_count":11},{"model":"Configurable-Hermes-2-Pro-Llama-3-8B","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-02","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":21,"benchmark_group_count":19,"benchmark_coverage_score":12.88,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":3.25,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.381,"source_primary_group_share":0.3684,"capability_breadth":3.22,"capability_primary":"knowledge_general","capability_primary_share":0.381,"capability_primary_group_count":7,"capability_non_primary_group_count":12},{"model":"calme-3.2-instruct-78B","provider":"Other","country":"XX","openness":"open","release_date":"2024-11-19","param_count":78.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":21,"benchmark_group_count":19,"benchmark_coverage_score":12.88,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.24,"source_primary":"huggingface.co","source_primary_share":0.5238,"source_primary_group_share":0.5789,"capability_breadth":3.22,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4762,"capability_primary_group_count":8,"capability_non_primary_group_count":11},{"model":"Qwen 3.7 Max (Max)","provider":"Alibaba","country":"CN","openness":"api","release_date":"2026-05-20","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":21,"benchmark_group_count":20,"benchmark_coverage_score":18.18,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.36,"source_primary":"benchmarklist.com","source_primary_share":0.85,"source_primary_group_share":0.85,"capability_breadth":4.55,"capability_primary":"math","capability_primary_share":0.35,"capability_primary_group_count":7,"capability_non_primary_group_count":13},{"model":"GPT-5 Codex","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-09-23","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":21,"benchmark_group_count":21,"benchmark_coverage_score":21.0,"source_family_count":16,"source_group_family_count":16,"source_effective_breadth":12.6,"source_primary":"benchmarklist.com","source_primary_share":0.1429,"source_primary_group_share":0.1429,"capability_breadth":5.73,"capability_primary":"code","capability_primary_share":0.3333,"capability_primary_group_count":7,"capability_non_primary_group_count":14},{"model":"Llama 3.2 90B","provider":"Meta","country":"US","openness":"open","release_date":"2024-09-25","param_count":90.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":21,"benchmark_group_count":11,"benchmark_coverage_score":11.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.57,"source_primary":"fab-ai.org","source_primary_share":0.6667,"source_primary_group_share":0.5455,"capability_breadth":3.0,"capability_primary":"vision_language","capability_primary_share":0.4286,"capability_primary_group_count":8,"capability_non_primary_group_count":3},{"model":"Claude Opus 4 (Non-reasoning)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-05-22","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 4.0 family deprecated in favor of newer Claude 4.x models.","num_benchmarks":21,"benchmark_group_count":20,"benchmark_coverage_score":20.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":3.45,"source_primary":"artificialanalysis.ai","source_primary_share":0.5,"source_primary_group_share":0.5,"capability_breadth":6.06,"capability_primary":"knowledge_general","capability_primary_share":0.3,"capability_primary_group_count":6,"capability_non_primary_group_count":14},{"model":"granite-7B-instruct","provider":"IBM","country":"US","openness":"open","release_date":"2024-05-19","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":21,"benchmark_group_count":16,"benchmark_coverage_score":13.47,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.88,"source_primary":"euroeval.com","source_primary_share":0.7143,"source_primary_group_share":0.625,"capability_breadth":3.37,"capability_primary":"knowledge_multilingual","capability_primary_share":0.381,"capability_primary_group_count":6,"capability_non_primary_group_count":10},{"model":"Grok 4.1","provider":"xAI","country":"US","openness":"api","release_date":"2025-11-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":20,"benchmark_group_count":18,"benchmark_coverage_score":18.0,"source_family_count":12,"source_group_family_count":12,"source_effective_breadth":9.53,"source_primary":"trackingai.org","source_primary_share":0.2,"source_primary_group_share":0.1667,"capability_breadth":5.26,"capability_primary":"code","capability_primary_share":0.25,"capability_primary_group_count":5,"capability_non_primary_group_count":13},{"model":"O3 (2025-04-16) (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-04-16","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":20,"benchmark_group_count":19,"benchmark_coverage_score":19.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":5.55,"source_primary":"epoch.ai","source_primary_share":0.2632,"source_primary_group_share":0.2632,"capability_breadth":5.55,"capability_primary":"knowledge_general","capability_primary_share":0.3158,"capability_primary_group_count":6,"capability_non_primary_group_count":13},{"model":"Phi-1","provider":"Microsoft","country":"US","openness":"open","release_date":"2023-06-20","param_count":1.3,"active_param_count":null,"training_compute_flop":3.323e+20,"num_benchmarks":20,"benchmark_group_count":15,"benchmark_coverage_score":13.75,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":4.41,"source_primary":"huggingface.co/spaces/aim-harvard/rabbits-leaderboard","source_primary_share":0.3684,"source_primary_group_share":0.4,"capability_breadth":3.44,"capability_primary":"knowledge_general","capability_primary_share":0.4737,"capability_primary_group_count":5,"capability_non_primary_group_count":10},{"model":"Qwen-7B Chat","provider":"Alibaba","country":"CN","openness":"open","release_date":"2023-08-01","param_count":7.0,"active_param_count":null,"training_compute_flop":1.01e+23,"num_benchmarks":20,"benchmark_group_count":17,"benchmark_coverage_score":17.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":3.75,"source_primary":"lk.instruct.kr","source_primary_share":0.45,"source_primary_group_share":0.4118,"capability_breadth":5.56,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3,"capability_primary_group_count":4,"capability_non_primary_group_count":13},{"model":"DBRX Instruct","provider":"Databricks","country":"US","openness":"open","release_date":"2024-03-27","param_count":132.0,"active_param_count":36.0,"training_compute_flop":2.6e+24,"num_benchmarks":20,"benchmark_group_count":19,"benchmark_coverage_score":13.56,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":3.25,"source_primary":"artificialanalysis.ai","source_primary_share":0.35,"source_primary_group_share":0.3684,"capability_breadth":3.39,"capability_primary":"knowledge_general","capability_primary_share":0.45,"capability_primary_group_count":8,"capability_non_primary_group_count":11},{"model":"Yi-34B-200K","provider":"01.AI","country":"CN","openness":"open","release_date":"2023-11-06","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":20,"benchmark_group_count":17,"benchmark_coverage_score":17.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":3.25,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.4444,"source_primary_group_share":0.4118,"capability_breadth":4.26,"capability_primary":"knowledge_general","capability_primary_share":0.3889,"capability_primary_group_count":6,"capability_non_primary_group_count":11},{"model":"NeuralSynthesis-7B-v0.4-slerp","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-12","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":20,"benchmark_group_count":18,"benchmark_coverage_score":11.94,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.95,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.4,"source_primary_group_share":0.3889,"capability_breadth":2.99,"capability_primary":"knowledge_general","capability_primary_share":0.4,"capability_primary_group_count":7,"capability_non_primary_group_count":11},{"model":"Nous-Hermes-2-SOLAR-10.7B","provider":"Nous Research","country":"US","openness":"open","release_date":"2024-01-01","param_count":10.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":20,"benchmark_group_count":16,"benchmark_coverage_score":14.29,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.91,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.4,"source_primary_group_share":0.4375,"capability_breadth":3.57,"capability_primary":"knowledge_general","capability_primary_share":0.35,"capability_primary_group_count":6,"capability_non_primary_group_count":10},{"model":"Gemini 3 Deep Think","provider":"Google","country":"US","openness":"api","release_date":"2026-02-26","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":20,"benchmark_group_count":18,"benchmark_coverage_score":18.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":3.52,"source_primary":"storage.googleapis.com","source_primary_share":0.45,"source_primary_group_share":0.5,"capability_breadth":4.65,"capability_primary":"math","capability_primary_share":0.3,"capability_primary_group_count":6,"capability_non_primary_group_count":12},{"model":"Jamba 1.6 Large","provider":"AI21 Labs","country":"IL","openness":"api","release_date":"2025-03-06","param_count":398.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":20,"benchmark_group_count":20,"benchmark_coverage_score":13.79,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.53,"source_primary":"vals.ai","source_primary_share":0.5,"source_primary_group_share":0.5,"capability_breadth":3.45,"capability_primary":"knowledge_general","capability_primary_share":0.45,"capability_primary_group_count":9,"capability_non_primary_group_count":11},{"model":"granite-7B-base","provider":"IBM","country":"US","openness":"open","release_date":"2024-04-19","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":20,"benchmark_group_count":16,"benchmark_coverage_score":13.56,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.88,"source_primary":"euroeval.com","source_primary_share":0.7,"source_primary_group_share":0.625,"capability_breadth":3.39,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4,"capability_primary_group_count":6,"capability_non_primary_group_count":10},{"model":"Yi-Coder-9B-Chat","provider":"01.AI","country":"CN","openness":"open","release_date":"2024-08-21","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":19,"benchmark_group_count":18,"benchmark_coverage_score":17.4,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.6,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.4211,"source_primary_group_share":0.3889,"capability_breadth":4.35,"capability_primary":"knowledge_general","capability_primary_share":0.3684,"capability_primary_group_count":6,"capability_non_primary_group_count":12},{"model":"Yi Large","provider":"01.AI","country":"CN","openness":"api","release_date":"2024-06-25","param_count":null,"active_param_count":null,"training_compute_flop":1.8e+24,"num_benchmarks":19,"benchmark_group_count":12,"benchmark_coverage_score":12.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":4.5,"source_primary":"fab-ai.org","source_primary_share":0.4211,"source_primary_group_share":0.4167,"capability_breadth":3.44,"capability_primary":"knowledge_general","capability_primary_share":0.4211,"capability_primary_group_count":6,"capability_non_primary_group_count":6},{"model":"chatglm3-6B","provider":"Zhipu","country":"CN","openness":"open","release_date":"2023-10-25","param_count":6.0,"active_param_count":null,"training_compute_flop":5.04e+22,"num_benchmarks":19,"benchmark_group_count":15,"benchmark_coverage_score":15.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.57,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.4211,"source_primary_group_share":0.4667,"capability_breadth":4.69,"capability_primary":"knowledge_general","capability_primary_share":0.3158,"capability_primary_group_count":5,"capability_non_primary_group_count":10},{"model":"Yi-9B","provider":"01.AI","country":"CN","openness":"open","release_date":"2024-03-01","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":19,"benchmark_group_count":18,"benchmark_coverage_score":12.78,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":3.18,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.4211,"source_primary_group_share":0.3889,"capability_breadth":3.19,"capability_primary":"knowledge_general","capability_primary_share":0.4737,"capability_primary_group_count":8,"capability_non_primary_group_count":10},{"model":"O1 (Medium)","provider":"OpenAI","country":"US","openness":"api","release_date":"2024-12-05","param_count":200.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":19,"benchmark_group_count":11,"benchmark_coverage_score":11.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":5.26,"source_primary":"fab-ai.org","source_primary_share":0.4211,"source_primary_group_share":0.3636,"capability_breadth":3.03,"capability_primary":"knowledge_general","capability_primary_share":0.4211,"capability_primary_group_count":5,"capability_non_primary_group_count":6},{"model":"Palmyra-X-004","provider":"Writer","country":"US","openness":"api","release_date":"2024-10-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":19,"benchmark_group_count":18,"benchmark_coverage_score":12.34,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.35,"source_primary":"crfm.stanford.edu","source_primary_share":0.5263,"source_primary_group_share":0.5556,"capability_breadth":3.09,"capability_primary":"knowledge_general","capability_primary_share":0.4211,"capability_primary_group_count":7,"capability_non_primary_group_count":11},{"model":"Gemini 1.0 Pro (002)","provider":"Google","country":"US","openness":"api","release_date":"2024-02-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 1.x family retired or superseded by newer Gemini models.","num_benchmarks":19,"benchmark_group_count":15,"benchmark_coverage_score":13.5,"source_family_count":6,"source_group_family_count":5,"source_effective_breadth":1.8,"source_primary":"crfm.stanford.edu","source_primary_share":0.7368,"source_primary_group_share":0.7333,"capability_breadth":3.37,"capability_primary":"knowledge_general","capability_primary_share":0.4211,"capability_primary_group_count":6,"capability_non_primary_group_count":9},{"model":"tulu-2-dpo-13B","provider":"Allen AI","country":"US","openness":"open","release_date":"2023-11-12","param_count":13.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":19,"benchmark_group_count":18,"benchmark_coverage_score":10.86,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":4.26,"source_primary":"huggingface.co/spaces/logikon/open_cot_leaderboard","source_primary_share":0.3158,"source_primary_group_share":0.3333,"capability_breadth":2.71,"capability_primary":"reasoning_logical","capability_primary_share":0.4737,"capability_primary_group_count":8,"capability_non_primary_group_count":10},{"model":"Llama-3-Instruct-8B-SimPO","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-17","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":19,"benchmark_group_count":18,"benchmark_coverage_score":17.83,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.6,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.4211,"source_primary_group_share":0.3889,"capability_breadth":4.46,"capability_primary":"knowledge_general","capability_primary_share":0.3684,"capability_primary_group_count":6,"capability_non_primary_group_count":12},{"model":"Gemini 1.5 Pro (Preview 0827)","provider":"Google","country":"US","openness":"api","release_date":"2024-08-27","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 1.x family retired or superseded by newer Gemini models.","num_benchmarks":19,"benchmark_group_count":19,"benchmark_coverage_score":19.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.24,"source_primary":"livebench.ai","source_primary_share":0.8947,"source_primary_group_share":0.8947,"capability_breadth":5.39,"capability_primary":"reasoning_logical","capability_primary_share":0.2632,"capability_primary_group_count":5,"capability_non_primary_group_count":14},{"model":"Gemini 1.5 Flash (Preview 0827)","provider":"Google","country":"US","openness":"api","release_date":"2024-08-27","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 1.x family retired or superseded by newer Gemini models.","num_benchmarks":19,"benchmark_group_count":19,"benchmark_coverage_score":19.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.24,"source_primary":"livebench.ai","source_primary_share":0.8947,"source_primary_group_share":0.8947,"capability_breadth":5.39,"capability_primary":"reasoning_logical","capability_primary_share":0.2632,"capability_primary_group_count":5,"capability_non_primary_group_count":14},{"model":"salamandra-2B","provider":"BSC","country":"ES","openness":"open","release_date":"2024-09-30","param_count":2.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":19,"benchmark_group_count":18,"benchmark_coverage_score":7.72,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.98,"source_primary":"euroeval.com","source_primary_share":0.5789,"source_primary_group_share":0.5556,"capability_breadth":1.93,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6842,"capability_primary_group_count":13,"capability_non_primary_group_count":5},{"model":"salamandra-2B-instruct","provider":"BSC","country":"ES","openness":"open","release_date":"2024-09-30","param_count":2.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":19,"benchmark_group_count":17,"benchmark_coverage_score":8.54,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"euroeval.com","source_primary_share":0.5789,"source_primary_group_share":0.5294,"capability_breadth":2.14,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6316,"capability_primary_group_count":12,"capability_non_primary_group_count":5},{"model":"maestrale-chat-v0.4-beta","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-06","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":19,"benchmark_group_count":14,"benchmark_coverage_score":10.1,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.56,"source_primary":"huggingface.co","source_primary_share":0.5789,"source_primary_group_share":0.7857,"capability_breadth":2.52,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5789,"capability_primary_group_count":10,"capability_non_primary_group_count":4},{"model":"Volare","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-15","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":19,"benchmark_group_count":14,"benchmark_coverage_score":10.1,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.56,"source_primary":"huggingface.co","source_primary_share":0.5789,"source_primary_group_share":0.7857,"capability_breadth":2.52,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5789,"capability_primary_group_count":10,"capability_non_primary_group_count":4},{"model":"occiglot-7B-it-en-instruct","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-05","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":19,"benchmark_group_count":14,"benchmark_coverage_score":10.1,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.56,"source_primary":"huggingface.co","source_primary_share":0.5789,"source_primary_group_share":0.7857,"capability_breadth":2.52,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5789,"capability_primary_group_count":10,"capability_non_primary_group_count":4},{"model":"Llama 3 8B 4bit UltraChat Ita","provider":"Meta","country":"US","openness":"open","release_date":"2024-04-18","param_count":8.0,"active_param_count":null,"training_compute_flop":7.2e+23,"num_benchmarks":19,"benchmark_group_count":14,"benchmark_coverage_score":10.1,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.56,"source_primary":"huggingface.co","source_primary_share":0.5789,"source_primary_group_share":0.7857,"capability_breadth":2.52,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5789,"capability_primary_group_count":10,"capability_non_primary_group_count":4},{"model":"LLaMAntino-2-7B-hf-ITA","provider":"Meta","country":"US","openness":"open","release_date":"2023-12-14","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":19,"benchmark_group_count":14,"benchmark_coverage_score":10.1,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.56,"source_primary":"huggingface.co","source_primary_share":0.5789,"source_primary_group_share":0.7857,"capability_breadth":2.52,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5789,"capability_primary_group_count":10,"capability_non_primary_group_count":4},{"model":"Claude Fable 5 (High)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-06-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":19,"benchmark_group_count":18,"benchmark_coverage_score":15.8,"source_family_count":11,"source_group_family_count":11,"source_effective_breadth":7.04,"source_primary":"arena.ai","source_primary_share":0.2778,"source_primary_group_share":0.2778,"capability_breadth":3.95,"capability_primary":"agentic","capability_primary_share":0.3889,"capability_primary_group_count":7,"capability_non_primary_group_count":11},{"model":"Palmyra X5","provider":"Writer","country":"US","openness":"api","release_date":"2025-04-28","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":19,"benchmark_group_count":19,"benchmark_coverage_score":15.53,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.6,"source_primary":"crfm.stanford.edu","source_primary_share":0.5263,"source_primary_group_share":0.5263,"capability_breadth":3.88,"capability_primary":"instruction_following","capability_primary_share":0.3684,"capability_primary_group_count":7,"capability_non_primary_group_count":12},{"model":"Jamba 1.6 Mini","provider":"AI21 Labs","country":"IL","openness":"api","release_date":"2025-03-06","param_count":52.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":19,"benchmark_group_count":19,"benchmark_coverage_score":12.56,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.36,"source_primary":"vals.ai","source_primary_share":0.5263,"source_primary_group_share":0.5263,"capability_breadth":3.14,"capability_primary":"knowledge_general","capability_primary_share":0.4737,"capability_primary_group_count":9,"capability_non_primary_group_count":10},{"model":"Gemini 2.5 Flash (Preview 09-2025) (Thinking)","provider":"Google","country":"US","openness":"api","release_date":"2025-09-25","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 2.5 preview snapshot retired or superseded by newer Gemini models.","num_benchmarks":19,"benchmark_group_count":18,"benchmark_coverage_score":10.98,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.38,"source_primary":"vals.ai","source_primary_share":0.8333,"source_primary_group_share":0.8333,"capability_breadth":2.75,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":10,"capability_non_primary_group_count":8},{"model":"MiniMax-Text-01","provider":"MiniMax","country":"CN","openness":"open","release_date":"2025-01-14","param_count":456.0,"active_param_count":45.9,"training_compute_flop":3.142e+24,"num_benchmarks":18,"benchmark_group_count":18,"benchmark_coverage_score":18.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":4.15,"source_primary":"prollm.ai","source_primary_share":0.4444,"source_primary_group_share":0.4444,"capability_breadth":6.0,"capability_primary":"reasoning_logical","capability_primary_share":0.2778,"capability_primary_group_count":5,"capability_non_primary_group_count":13},{"model":"Mistral Small","provider":"Mistral","country":"FR","openness":"open","release_date":"2025-01-30","param_count":22.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":18,"benchmark_group_count":18,"benchmark_coverage_score":18.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":4.76,"source_primary":"prollm.ai","source_primary_share":0.3889,"source_primary_group_share":0.3889,"capability_breadth":5.4,"capability_primary":"code","capability_primary_share":0.3333,"capability_primary_group_count":6,"capability_non_primary_group_count":12},{"model":"OpenCoder-8B-Instruct","provider":"Other","country":"XX","openness":"open","release_date":"2024-11-07","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":18,"benchmark_group_count":16,"benchmark_coverage_score":11.78,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":3.05,"source_primary":"euroeval.com","source_primary_share":0.5556,"source_primary_group_share":0.5,"capability_breadth":2.95,"capability_primary":"code","capability_primary_share":0.4444,"capability_primary_group_count":8,"capability_non_primary_group_count":8},{"model":"Qwen-14B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2023-09-25","param_count":14.0,"active_param_count":null,"training_compute_flop":2.5e+23,"num_benchmarks":18,"benchmark_group_count":12,"benchmark_coverage_score":12.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.4,"source_primary":"epoch.ai","source_primary_share":0.4444,"source_primary_group_share":0.5833,"capability_breadth":3.95,"capability_primary":"code","capability_primary_share":0.4444,"capability_primary_group_count":3,"capability_non_primary_group_count":9},{"model":"deepseek-llm-7B-chat","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2023-11-29","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":18,"benchmark_group_count":17,"benchmark_coverage_score":17.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":3.25,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.4444,"source_primary_group_share":0.4118,"capability_breadth":4.38,"capability_primary":"knowledge_general","capability_primary_share":0.3889,"capability_primary_group_count":6,"capability_non_primary_group_count":11},{"model":"O3 Mini (Low)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-01-31","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":18,"benchmark_group_count":17,"benchmark_coverage_score":17.0,"source_family_count":16,"source_group_family_count":16,"source_effective_breadth":15.21,"source_primary":"arcprize.org","source_primary_share":0.1111,"source_primary_group_share":0.1176,"capability_breadth":5.59,"capability_primary":"math","capability_primary_share":0.2778,"capability_primary_group_count":4,"capability_non_primary_group_count":13},{"model":"Sonar Pro","provider":"Perplexity","country":"US","openness":"api","release_date":"2025-02-01","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":18,"benchmark_group_count":13,"benchmark_coverage_score":13.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.77,"source_primary":"artificialanalysis.ai","source_primary_share":0.4375,"source_primary_group_share":0.5385,"capability_breadth":4.57,"capability_primary":"knowledge_general","capability_primary_share":0.3125,"capability_primary_group_count":5,"capability_non_primary_group_count":8},{"model":"Qwen 1.5 110B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-04-25","param_count":110.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":18,"benchmark_group_count":17,"benchmark_coverage_score":15.8,"source_family_count":8,"source_group_family_count":7,"source_effective_breadth":4.19,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.3333,"source_primary_group_share":0.3529,"capability_breadth":3.95,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":7,"capability_non_primary_group_count":10},{"model":"Llama-PLLuM-8B-chat","provider":"Meta","country":"US","openness":"open","release_date":"2025-02-24","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":18,"benchmark_group_count":18,"benchmark_coverage_score":12.96,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.7,"source_primary":"huggingface.co/spaces/speakleash/mt-bench-pl","source_primary_share":0.5,"source_primary_group_share":0.5,"capability_breadth":3.24,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5,"capability_primary_group_count":9,"capability_non_primary_group_count":9},{"model":"Yi-1.5-34B-Chat-16K","provider":"01.AI","country":"CN","openness":"open","release_date":"2024-05-15","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":18,"benchmark_group_count":14,"benchmark_coverage_score":9.53,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.4444,"source_primary_group_share":0.5,"capability_breadth":2.38,"capability_primary":"knowledge_general","capability_primary_share":0.6111,"capability_primary_group_count":7,"capability_non_primary_group_count":7},{"model":"Sonar","provider":"Perplexity","country":"US","openness":"api","release_date":"2025-02-01","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":18,"benchmark_group_count":11,"benchmark_coverage_score":8.2,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.28,"source_primary":"fab-ai.org","source_primary_share":0.4444,"source_primary_group_share":0.6364,"capability_breadth":2.05,"capability_primary":"knowledge_general","capability_primary_share":0.6667,"capability_primary_group_count":6,"capability_non_primary_group_count":5},{"model":"Gemma 4 26B","provider":"Google","country":"US","openness":"api","release_date":"2026-04-02","param_count":26.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":18,"benchmark_group_count":17,"benchmark_coverage_score":9.82,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.66,"source_primary":"benchmarklist.com","source_primary_share":0.3333,"source_primary_group_share":0.3529,"capability_breadth":2.45,"capability_primary":"reasoning_logical","capability_primary_share":0.6111,"capability_primary_group_count":10,"capability_non_primary_group_count":7},{"model":"O3 Pro","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-06-10","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":18,"benchmark_group_count":14,"benchmark_coverage_score":14.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":6.13,"source_primary":"benchmarklist.com","source_primary_share":0.2353,"source_primary_group_share":0.2857,"capability_breadth":5.07,"capability_primary":"knowledge_general","capability_primary_share":0.2353,"capability_primary_group_count":4,"capability_non_primary_group_count":10},{"model":"phi-4-14B","provider":"Microsoft","country":"US","openness":"open","release_date":"2025-01-16","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":18,"benchmark_group_count":16,"benchmark_coverage_score":15.52,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.46,"source_primary":"huggingface.co/spaces/gsma/open-telco-llm-leaderboard","source_primary_share":0.5,"source_primary_group_share":0.5,"capability_breadth":3.88,"capability_primary":"knowledge_general","capability_primary_share":0.4375,"capability_primary_group_count":7,"capability_non_primary_group_count":9},{"model":"GPT-5.1 Codex Mini","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-11-19","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":18,"benchmark_group_count":16,"benchmark_coverage_score":16.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":6.1,"source_primary":"designarena.ai","source_primary_share":0.3333,"source_primary_group_share":0.25,"capability_breadth":6.0,"capability_primary":"code","capability_primary_share":0.3333,"capability_primary_group_count":5,"capability_non_primary_group_count":11},{"model":"Gemma 4 12B","provider":"Google","country":"US","openness":"api","release_date":"2026-04-02","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":18,"benchmark_group_count":16,"benchmark_coverage_score":16.0,"source_family_count":6,"source_group_family_count":5,"source_effective_breadth":2.91,"source_primary":"llm-stats.com","source_primary_share":0.4444,"source_primary_group_share":0.4375,"capability_breadth":6.23,"capability_primary":"vision_language","capability_primary_share":0.2222,"capability_primary_group_count":4,"capability_non_primary_group_count":12},{"model":"Claude Opus 4.8 (High)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-05-29","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":18,"benchmark_group_count":18,"benchmark_coverage_score":18.0,"source_family_count":13,"source_group_family_count":13,"source_effective_breadth":10.8,"source_primary":"arcprize.org","source_primary_share":0.1667,"source_primary_group_share":0.1667,"capability_breadth":5.79,"capability_primary":"code","capability_primary_share":0.2778,"capability_primary_group_count":5,"capability_non_primary_group_count":13},{"model":"Llama 3.1 Nemotron Ultra 253B v1","provider":"NVIDIA","country":"US","openness":"open","release_date":"2025-04-07","param_count":253.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":18,"benchmark_group_count":16,"benchmark_coverage_score":16.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":7.53,"source_primary":"dubesor.de","source_primary_share":0.1765,"source_primary_group_share":0.1875,"capability_breadth":6.42,"capability_primary":"knowledge_general","capability_primary_share":0.2941,"capability_primary_group_count":5,"capability_non_primary_group_count":11},{"model":"gpt-sw3-1.3B-instruct","provider":"AI Sweden","country":"SE","openness":"open","release_date":"2023-04-28","param_count":1.3,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":18,"benchmark_group_count":12,"benchmark_coverage_score":7.12,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.95,"source_primary":"euroeval.com","source_primary_share":0.6111,"source_primary_group_share":0.5833,"capability_breadth":1.78,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7222,"capability_primary_group_count":9,"capability_non_primary_group_count":3},{"model":"ERNIE 5.0","provider":"Baidu","country":"CN","openness":"api","release_date":"2025-11-13","param_count":2400.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":17,"benchmark_group_count":16,"benchmark_coverage_score":12.43,"source_family_count":10,"source_group_family_count":10,"source_effective_breadth":6.1,"source_primary":"raw.githubusercontent.com/llm2014/llm_benchmark","source_primary_share":0.3529,"source_primary_group_share":0.3125,"capability_breadth":3.11,"capability_primary":"reasoning_logical","capability_primary_share":0.5294,"capability_primary_group_count":8,"capability_non_primary_group_count":8},{"model":"Claude Sonnet 4.5 (Thinking 32K)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-09-29","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":17,"benchmark_group_count":17,"benchmark_coverage_score":16.28,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.98,"source_primary":"epoch.ai","source_primary_share":0.5294,"source_primary_group_share":0.5294,"capability_breadth":4.07,"capability_primary":"math","capability_primary_share":0.4118,"capability_primary_group_count":7,"capability_non_primary_group_count":10},{"model":"DeepSeek Reasoner","provider":"DeepSeek","country":"CN","openness":"api","release_date":"2025-01-20","param_count":671.0,"active_param_count":37.0,"training_compute_flop":null,"num_benchmarks":17,"benchmark_group_count":14,"benchmark_coverage_score":14.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":5.76,"source_primary":"designarena.ai","source_primary_share":0.375,"source_primary_group_share":0.2857,"capability_breadth":6.1,"capability_primary":"code","capability_primary_share":0.3125,"capability_primary_group_count":4,"capability_non_primary_group_count":10},{"model":"llama-3.2-Korean-Bllossom-AICA-5B","provider":"Meta","country":"US","openness":"open","release_date":"2024-12-12","param_count":5.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":17,"benchmark_group_count":17,"benchmark_coverage_score":8.82,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.84,"source_primary":"huggingface.co/spaces/maum-ai/koffvqa-leaderboard","source_primary_share":0.6471,"source_primary_group_share":0.6471,"capability_breadth":2.21,"capability_primary":"vision_language","capability_primary_share":0.6471,"capability_primary_group_count":11,"capability_non_primary_group_count":6},{"model":"tulu-2-7B","provider":"Allen AI","country":"US","openness":"open","release_date":"2023-11-13","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":17,"benchmark_group_count":15,"benchmark_coverage_score":13.13,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.47,"source_primary":"huggingface.co/spaces/ai-secure/llm-trustworthy-leaderboard","source_primary_share":0.5,"source_primary_group_share":0.5333,"capability_breadth":3.28,"capability_primary":"reasoning_logical","capability_primary_share":0.375,"capability_primary_group_count":5,"capability_non_primary_group_count":10},{"model":"Claude Instant","provider":"Anthropic","country":"US","openness":"api","release_date":"2023-03-14","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 1/2 family retired or superseded by newer Claude models.","num_benchmarks":17,"benchmark_group_count":15,"benchmark_coverage_score":15.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.23,"source_primary":"huggingface.co/spaces/nphardeval/nphardeval-leaderboard","source_primary_share":0.5882,"source_primary_group_share":0.5333,"capability_breadth":4.07,"capability_primary":"reasoning_logical","capability_primary_share":0.2941,"capability_primary_group_count":4,"capability_non_primary_group_count":11},{"model":"gemma-2-9B-it-SimPO","provider":"Google","country":"US","openness":"api","release_date":"2024-07-16","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":17,"benchmark_group_count":16,"benchmark_coverage_score":15.01,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.88,"source_primary":"huggingface.co/spaces/logikon/open_cot_leaderboard","source_primary_share":0.3529,"source_primary_group_share":0.375,"capability_breadth":3.75,"capability_primary":"reasoning_logical","capability_primary_share":0.4118,"capability_primary_group_count":6,"capability_non_primary_group_count":10},{"model":"Infinity-Instruct-3M-0625-Llama3-8B","provider":"BAAI","country":"CN","openness":"open","release_date":"2024-07-09","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":17,"benchmark_group_count":16,"benchmark_coverage_score":16.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.91,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.4706,"source_primary_group_share":0.4375,"capability_breadth":4.07,"capability_primary":"knowledge_general","capability_primary_share":0.4118,"capability_primary_group_count":6,"capability_non_primary_group_count":10},{"model":"Liberated-Qwen1.5-14B","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-01","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":17,"benchmark_group_count":9,"benchmark_coverage_score":7.66,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.98,"source_primary":"huggingface.co/spaces/openlifescienceai/open_medical_llm_leaderboard","source_primary_share":0.5882,"source_primary_group_share":0.6667,"capability_breadth":1.91,"capability_primary":"knowledge_general","capability_primary_share":0.7059,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Seed 1.8","provider":"ByteDance","country":"CN","openness":"api","release_date":"2025-12-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":17,"benchmark_group_count":17,"benchmark_coverage_score":17.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":3.85,"source_primary":"benchmarklist.com","source_primary_share":0.4706,"source_primary_group_share":0.4706,"capability_breadth":4.31,"capability_primary":"vision_language","capability_primary_share":0.4118,"capability_primary_group_count":7,"capability_non_primary_group_count":10},{"model":"Claude Opus 4.7 (Max)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-04-16","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":17,"benchmark_group_count":17,"benchmark_coverage_score":17.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":3.25,"source_primary":"epoch.ai","source_primary_share":0.4118,"source_primary_group_share":0.4118,"capability_breadth":4.74,"capability_primary":"math","capability_primary_share":0.2941,"capability_primary_group_count":5,"capability_non_primary_group_count":12},{"model":"Skywork-Reward-Gemma-2-27B-v0.2","provider":"Skywork","country":"CN","openness":"open","release_date":"2024-10-14","param_count":27.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":17,"benchmark_group_count":17,"benchmark_coverage_score":17.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.11,"source_primary":"huggingface.co","source_primary_share":0.5882,"source_primary_group_share":0.5882,"capability_breadth":4.31,"capability_primary":"reasoning_logical","capability_primary_share":0.3529,"capability_primary_group_count":6,"capability_non_primary_group_count":11},{"model":"Mistral Medium 3.5 (High)","provider":"Mistral","country":"FR","openness":"api","release_date":"2025-12-02","param_count":123.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":17,"benchmark_group_count":16,"benchmark_coverage_score":16.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":6.1,"source_primary":"benchmarklist.com","source_primary_share":0.3529,"source_primary_group_share":0.3125,"capability_breadth":7.81,"capability_primary":"agentic","capability_primary_share":0.1765,"capability_primary_group_count":3,"capability_non_primary_group_count":13},{"model":"openchat-3.5-1210","provider":"Other","country":"XX","openness":"open","release_date":"2023-12-12","param_count":7.2,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":17,"benchmark_group_count":14,"benchmark_coverage_score":14.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.65,"source_primary":"artificialanalysis.ai","source_primary_share":0.3529,"source_primary_group_share":0.4286,"capability_breadth":4.59,"capability_primary":"knowledge_general","capability_primary_share":0.2941,"capability_primary_group_count":5,"capability_non_primary_group_count":9},{"model":"SmolLM-1.7B-Instruct","provider":"HuggingFace","country":"FR","openness":"open","release_date":"2024-07-15","param_count":1.7,"active_param_count":null,"training_compute_flop":1.03e+22,"num_benchmarks":17,"benchmark_group_count":13,"benchmark_coverage_score":13.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.68,"source_primary":"huggingface.co/spaces/logikon/open_cot_leaderboard","source_primary_share":0.3529,"source_primary_group_share":0.4615,"capability_breadth":3.75,"capability_primary":"reasoning_logical","capability_primary_share":0.4118,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"SmolLM-135M-Instruct","provider":"HuggingFace","country":"FR","openness":"open","release_date":"2024-07-15","param_count":0.135,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":17,"benchmark_group_count":13,"benchmark_coverage_score":13.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.68,"source_primary":"huggingface.co/spaces/logikon/open_cot_leaderboard","source_primary_share":0.3529,"source_primary_group_share":0.4615,"capability_breadth":3.75,"capability_primary":"reasoning_logical","capability_primary_share":0.4118,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Claude Opus 4.8 (Thinking)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-05-29","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":17,"benchmark_group_count":15,"benchmark_coverage_score":15.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.08,"source_primary":"arena.ai","source_primary_share":0.5625,"source_primary_group_share":0.5333,"capability_breadth":4.92,"capability_primary":"agentic","capability_primary_share":0.3125,"capability_primary_group_count":5,"capability_non_primary_group_count":10},{"model":"O1 (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2024-12-05","param_count":200.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":16,"benchmark_group_count":15,"benchmark_coverage_score":13.47,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":5.77,"source_primary":"fab-ai.org","source_primary_share":0.375,"source_primary_group_share":0.3333,"capability_breadth":3.37,"capability_primary":"vision_language","capability_primary_share":0.4375,"capability_primary_group_count":6,"capability_non_primary_group_count":9},{"model":"Athene-V2-Chat","provider":"Nexusflow","country":"US","openness":"open","release_date":"2024-11-14","param_count":72.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":16,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":4.0,"source_primary":"huggingface.co/spaces/toloka/u-math-leaderboard","source_primary_share":0.4375,"source_primary_group_share":0.375,"capability_breadth":3.05,"capability_primary":"math","capability_primary_share":0.4375,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"deepseek-llm-7B-base","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2023-11-29","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":16,"benchmark_group_count":15,"benchmark_coverage_score":12.8,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.59,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5,"source_primary_group_share":0.4667,"capability_breadth":3.2,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":7,"capability_non_primary_group_count":8},{"model":"tulu-2-13B","provider":"Allen AI","country":"US","openness":"open","release_date":"2023-11-13","param_count":13.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":16,"benchmark_group_count":14,"benchmark_coverage_score":13.04,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.18,"source_primary":"huggingface.co/spaces/ai-secure/llm-trustworthy-leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5714,"capability_breadth":3.26,"capability_primary":"reasoning_logical","capability_primary_share":0.4,"capability_primary_group_count":5,"capability_non_primary_group_count":9},{"model":"Yi-9B-200K","provider":"01.AI","country":"CN","openness":"open","release_date":"2024-03-15","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":16,"benchmark_group_count":15,"benchmark_coverage_score":15.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.59,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5,"source_primary_group_share":0.4667,"capability_breadth":3.76,"capability_primary":"knowledge_general","capability_primary_share":0.4375,"capability_primary_group_count":6,"capability_non_primary_group_count":9},{"model":"Infinity-Instruct-7M-Gen-Llama3.1-8B","provider":"BAAI","country":"CN","openness":"open","release_date":"2024-08-02","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":16,"benchmark_group_count":15,"benchmark_coverage_score":14.63,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.59,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5,"source_primary_group_share":0.4667,"capability_breadth":3.66,"capability_primary":"knowledge_general","capability_primary_share":0.4375,"capability_primary_group_count":6,"capability_non_primary_group_count":9},{"model":"Infinity-Instruct-3M-0625-Yi-1.5-9B","provider":"BAAI","country":"CN","openness":"open","release_date":"2024-07-09","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":16,"benchmark_group_count":15,"benchmark_coverage_score":14.63,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.59,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5,"source_primary_group_share":0.4667,"capability_breadth":3.66,"capability_primary":"knowledge_general","capability_primary_share":0.4375,"capability_primary_group_count":6,"capability_non_primary_group_count":9},{"model":"Hermes-3-Llama-3.2-3B","provider":"Nous Research","country":"US","openness":"open","release_date":"2024-12-03","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":16,"benchmark_group_count":16,"benchmark_coverage_score":15.06,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.17,"source_primary":"huggingface.co/spaces/speakleash/mt-bench-pl","source_primary_share":0.5625,"source_primary_group_share":0.5625,"capability_breadth":3.76,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4375,"capability_primary_group_count":7,"capability_non_primary_group_count":9},{"model":"Llama-3-Instruct-8B-SPPO-Iter3","provider":"Meta","country":"US","openness":"open","release_date":"2024-06-25","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":16,"benchmark_group_count":15,"benchmark_coverage_score":14.22,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.59,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5,"source_primary_group_share":0.4667,"capability_breadth":3.56,"capability_primary":"knowledge_general","capability_primary_share":0.4375,"capability_primary_group_count":6,"capability_non_primary_group_count":9},{"model":"suzume-llama-3-8B-multilingual-orpo-borda-half","provider":"Lightblue","country":"JP","openness":"open","release_date":"2024-04-25","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":16,"benchmark_group_count":15,"benchmark_coverage_score":14.22,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.59,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5,"source_primary_group_share":0.4667,"capability_breadth":3.56,"capability_primary":"knowledge_general","capability_primary_share":0.4375,"capability_primary_group_count":6,"capability_non_primary_group_count":9},{"model":"Mistral-7B-OpenOrca","provider":"Mistral","country":"FR","openness":"open","release_date":"2023-09-29","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":16,"benchmark_group_count":15,"benchmark_coverage_score":14.75,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.23,"source_primary":"huggingface.co/spaces/ai-secure/llm-trustworthy-leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5333,"capability_breadth":3.69,"capability_primary":"instruction_following","capability_primary_share":0.3333,"capability_primary_group_count":5,"capability_non_primary_group_count":10},{"model":"Yi Lightning","provider":"01.AI","country":"CN","openness":"api","release_date":"2024-10-16","param_count":null,"active_param_count":null,"training_compute_flop":1.5e+24,"num_benchmarks":16,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":5.4,"source_primary":"fab-ai.org","source_primary_share":0.5,"source_primary_group_share":0.3333,"capability_breadth":2.72,"capability_primary":"knowledge_general","capability_primary_share":0.5625,"capability_primary_group_count":3,"capability_non_primary_group_count":6},{"model":"llama-3-Korean-Bllossom-8B","provider":"Meta","country":"US","openness":"open","release_date":"2024-04-25","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":16,"benchmark_group_count":14,"benchmark_coverage_score":14.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"lk.instruct.kr","source_primary_share":0.5625,"source_primary_group_share":0.5,"capability_breadth":3.76,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4375,"capability_primary_group_count":5,"capability_non_primary_group_count":9},{"model":"nox-solar-10.7B-v4","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-16","param_count":10.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":16,"benchmark_group_count":14,"benchmark_coverage_score":14.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"lk.instruct.kr","source_primary_share":0.5625,"source_primary_group_share":0.5,"capability_breadth":3.76,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4375,"capability_primary_group_count":5,"capability_non_primary_group_count":9},{"model":"WizardLM-2-7B","provider":"Microsoft","country":"US","openness":"open","release_date":"2024-04-16","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":16,"benchmark_group_count":13,"benchmark_coverage_score":13.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.07,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.375,"source_primary_group_share":0.4615,"capability_breadth":4.92,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3125,"capability_primary_group_count":3,"capability_non_primary_group_count":10},{"model":"Claude Sonnet 4.6 (Max)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-02-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":16,"benchmark_group_count":15,"benchmark_coverage_score":15.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.65,"source_primary":"benchmarklist.com","source_primary_share":0.5625,"source_primary_group_share":0.5333,"capability_breadth":5.57,"capability_primary":"agentic","capability_primary_share":0.25,"capability_primary_group_count":4,"capability_non_primary_group_count":11},{"model":"Qwen 3 Next 80B A3B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-09-11","param_count":80.0,"active_param_count":3.0,"training_compute_flop":2.7e+23,"num_benchmarks":16,"benchmark_group_count":13,"benchmark_coverage_score":13.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.31,"source_primary":"huggingface.co/spaces/huggingface-krew/ko-agentbench","source_primary_share":0.375,"source_primary_group_share":0.4615,"capability_breadth":5.33,"capability_primary":"longctx","capability_primary_share":0.25,"capability_primary_group_count":4,"capability_non_primary_group_count":9},{"model":"command-a-reasoning-08-2025","provider":"Cohere","country":"CA","openness":"open","release_date":"2025-08-21","param_count":111.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":16,"benchmark_group_count":12,"benchmark_coverage_score":12.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.67,"source_primary":"huggingface.co","source_primary_share":0.5625,"source_primary_group_share":0.75,"capability_breadth":3.37,"capability_primary":"math","capability_primary_share":0.5,"capability_primary_group_count":4,"capability_non_primary_group_count":8},{"model":"Qwen1.5 Chat (72B)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-02-04","param_count":72.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":16,"benchmark_group_count":13,"benchmark_coverage_score":13.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.32,"source_primary":"crfm.stanford.edu","source_primary_share":0.5625,"source_primary_group_share":0.4615,"capability_breadth":3.37,"capability_primary":"instruction_following","capability_primary_share":0.3125,"capability_primary_group_count":5,"capability_non_primary_group_count":8},{"model":"Llama 3 8B SFR Iterative DPO R","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-09","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":16,"benchmark_group_count":15,"benchmark_coverage_score":11.64,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.92,"source_primary":"huggingface.co/spaces/logikon/open_cot_leaderboard","source_primary_share":0.375,"source_primary_group_share":0.4,"capability_breadth":2.91,"capability_primary":"reasoning_logical","capability_primary_share":0.4375,"capability_primary_group_count":6,"capability_non_primary_group_count":9},{"model":"Mistral-Small-24B-Base-2501","provider":"Mistral","country":"FR","openness":"open","release_date":"2025-01-23","param_count":24.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":16,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"huggingface.co/spaces/yangzhang33/greekmmlu-leaderboard","source_primary_share":0.375,"source_primary_group_share":0.75,"capability_breadth":2.67,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5625,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"Claude Sonnet 5 (High)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-06-30","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":16,"benchmark_group_count":14,"benchmark_coverage_score":14.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.8,"source_primary":"arena.ai","source_primary_share":0.6,"source_primary_group_share":0.5714,"capability_breadth":4.41,"capability_primary":"agentic","capability_primary_share":0.3333,"capability_primary_group_count":5,"capability_non_primary_group_count":9},{"model":"T-lite-instruct-0.1","provider":"Other","country":"XX","openness":"open","release_date":null,"param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":15,"benchmark_coverage_score":10.84,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.27,"source_primary":"huggingface.co/spaces/ai-forever/libra-leaderboard","source_primary_share":0.6,"source_primary_group_share":0.6,"capability_breadth":2.71,"capability_primary":"longctx","capability_primary_share":0.5333,"capability_primary_group_count":8,"capability_non_primary_group_count":7},{"model":"speechless-code-mistral-7B-v1.0","provider":"Other","country":"XX","openness":"open","release_date":"2023-10-10","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.85,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.46,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Vicuna-7B","provider":"LMSYS","country":"US","openness":"open","release_date":"2023-04-05","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.85,"source_family_count":8,"source_group_family_count":7,"source_effective_breadth":3.38,"source_primary":"allenai.github.io","source_primary_share":0.4667,"source_primary_group_share":0.5,"capability_breadth":3.46,"capability_primary":"instruction_following","capability_primary_share":0.4667,"capability_primary_group_count":7,"capability_non_primary_group_count":7},{"model":"Llama 3 70B Synthia V3.5","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-26","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.85,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.46,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"ReflectionCoder-DS-33B","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-28","param_count":33.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.85,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.46,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"ReflectionCoder-CL-34B","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-28","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.85,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.46,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"internlm2.5-7B-chat","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2024-06-27","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":12,"benchmark_coverage_score":12.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.43,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.4,"source_primary_group_share":0.5,"capability_breadth":4.79,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3333,"capability_primary_group_count":3,"capability_non_primary_group_count":9},{"model":"text-davinci-001","provider":"OpenAI","country":"US","openness":"api","release_date":"2022-01-27","param_count":175.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI completions family retired or superseded by newer API models.","num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.52,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"epoch.ai","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":3.38,"capability_primary":"knowledge_general","capability_primary_share":0.4286,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Llama3-ChatQA-1.5-8B","provider":"Meta","country":"US","openness":"open","release_date":"2024-04-28","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":11,"benchmark_coverage_score":10.34,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.6364,"capability_breadth":2.59,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4667,"capability_primary_group_count":5,"capability_non_primary_group_count":6},{"model":"luxia-21.4B-alignment-v1.0","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-12","param_count":21.4,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"luxia-21.4B-alignment-v1.2","provider":"Other","country":"XX","openness":"open","release_date":"2024-10-25","param_count":21.4,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Barcenas-14B-Phi-3-medium-ORPO","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-15","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"SauerkrautLM-Phi-3 (Medium)","provider":"VAGOsolutions","country":"DE","openness":"open","release_date":"2024-06-09","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"dolphin-2.9.1-yi-1.5-9B","provider":"Cognitive Computations","country":"US","openness":"open","release_date":"2024-05-18","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Yi-1.5-9B-Chat-16K-abliterated","provider":"01.AI","country":"CN","openness":"open","release_date":"2024-09-03","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"CarbonBeagle-11B","provider":"Other","country":"XX","openness":"open","release_date":"2024-01-21","param_count":11.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"free-solar-evo-v0.11","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-24","param_count":10.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"free-solar-evo-v0.13","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-28","param_count":10.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Yi-1.5-9B-Chat-abliterated","provider":"01.AI","country":"CN","openness":"open","release_date":"2024-09-04","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Llama 3 8B Instruct V0.10","provider":"Meta","country":"US","openness":"open","release_date":"2024-06-04","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"free-solar-evo-v0.1","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-18","param_count":10.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Llama 3 8B Instruct V0.9","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-30","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"ConfigurableBeagle-11B","provider":"Other","country":"XX","openness":"open","release_date":"2024-02-17","param_count":11.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Llama 3 8B Instruct V0.8","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-01","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"NeuralSynthesis-7B-v0.1","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-06","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"dzakwan-MoE-4x7B-Beta","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-11","param_count":28.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Beyonder-4x7B-v3","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-21","param_count":28.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Mistral-Small-Instruct-2409-abliterated","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-09-23","param_count":22.2,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"MixTAO-7Bx2-MoE-v8.1","provider":"Other","country":"XX","openness":"open","release_date":"2024-02-26","param_count":12.9,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"bigyi-15B","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-06","param_count":15.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Calme-4x7B-MoE-v0.1","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-17","param_count":28.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Calme-4x7B-MoE-v0.2","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-17","param_count":28.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"MAmmoTH2-7B-Plus","provider":"CMU","country":"US","openness":"open","release_date":"2024-05-06","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"dolphin-2.9.2-Phi-3-Medium-abliterated","provider":"Cognitive Computations","country":"US","openness":"open","release_date":"2024-06-03","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"dolphin-2.9.2-Phi-3 (Medium)","provider":"Cognitive Computations","country":"US","openness":"open","release_date":"2024-05-31","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Llama 3 8B Magpie Align V0.3","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-15","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Llama-3-Instruct-8B-SPPO-Iter3-SimPO-merge","provider":"Meta","country":"US","openness":"open","release_date":"2024-06-28","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Llama-3-Instruct-8B-SimPO-SPPO-Iter3-merge","provider":"Meta","country":"US","openness":"open","release_date":"2024-06-28","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Llama 3 8B ProLong 512k Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2024-08-22","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Llama 3 8B ProLong 64k Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-21","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Llama-3-Instruct-8B-CPO-SimPO","provider":"Meta","country":"US","openness":"open","release_date":"2024-06-19","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"suzume-llama-3-8B-multilingual-orpo-borda-full","provider":"Lightblue","country":"JP","openness":"open","release_date":"2024-04-25","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"suzume-llama-3-8B-multilingual-orpo-borda-top75","provider":"Lightblue","country":"JP","openness":"open","release_date":"2024-04-26","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Llama-3-Instruct-8B-SPPO-Iter2","provider":"Meta","country":"US","openness":"open","release_date":"2024-06-25","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Infinity-Instruct-7M-0729-Llama3.1-8B","provider":"BAAI","country":"CN","openness":"open","release_date":"2024-08-02","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"SeaLLM-7B-v2","provider":"Sea AI Lab","country":"SG","openness":"open","release_date":"2024-01-29","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"suzume-llama-3-8B-multilingual-orpo-borda-top25","provider":"Lightblue","country":"JP","openness":"open","release_date":"2024-04-26","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Mistral7B-PairRM-SPPO-ExPO","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-05-04","param_count":7.2,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Llama 3 8B Synthia V3.5","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-17","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Llama 3 8B Magpie Align SFT V0.3","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-13","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"mistral-orpo-capybara-7k","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-03-23","param_count":7.2,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Mistral7B-PairRM-SPPO-Iter2","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-05-04","param_count":7.2,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Mistral7B-PairRM-SPPO-Iter3","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-05-04","param_count":7.2,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Configurable-Janus-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-13","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":13,"benchmark_coverage_score":11.69,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.25,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5385,"capability_breadth":2.92,"capability_primary":"knowledge_general","capability_primary_share":0.4,"capability_primary_group_count":5,"capability_non_primary_group_count":8},{"model":"GritLM-7B-KTO","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-16","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"ghost-8B-beta-1608","provider":"Other","country":"XX","openness":"open","release_date":"2024-08-18","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"bigstral-12B-32k","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-06","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Llama-3-Swallow-8B-Instruct-v0.1","provider":"Meta","country":"US","openness":"open","release_date":"2024-06-26","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"rombos_Replete-Coder-Instruct-8B-Merged","provider":"Other","country":"XX","openness":"open","release_date":"2024-10-06","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"ConfigurableHermes-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-02-17","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Llama 3 8B ProLong 64k Base","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-22","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Llama 3 8B Orpo V0.1","provider":"Meta","country":"US","openness":"open","release_date":"2024-04-26","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"CodeNinja-1.0-OpenChat-7B","provider":"Other","country":"XX","openness":"open","release_date":"2023-12-20","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Llama 3 8B ProLong 512k Base","provider":"Meta","country":"US","openness":"open","release_date":"2024-08-22","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"rombos_Replete-Coder-Llama3-8B","provider":"Other","country":"XX","openness":"open","release_date":"2024-10-06","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"MegaBeam-Mistral-7B-512k","provider":"Other","country":"XX","openness":"open","release_date":"2024-07-30","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"MegaBeam-Mistral-7B-300k","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-13","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.43,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.36,"capability_primary":"knowledge_general","capability_primary_share":0.4667,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"Llama-VARCO-8B-Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2024-09-12","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":13,"benchmark_coverage_score":13.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"lk.instruct.kr","source_primary_share":0.6,"source_primary_group_share":0.5385,"capability_breadth":4.09,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4,"capability_primary_group_count":4,"capability_non_primary_group_count":9},{"model":"GLM-4 32B","provider":"Zhipu","country":"CN","openness":"open","release_date":"2025-04-14","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":13,"benchmark_coverage_score":13.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.76,"source_primary":"benchmarklist.com","source_primary_share":0.4615,"source_primary_group_share":0.4615,"capability_breadth":7.35,"capability_primary":"reasoning_logical","capability_primary_share":0.2308,"capability_primary_group_count":3,"capability_non_primary_group_count":10},{"model":"Nova 2 Lite","provider":"Amazon","country":"US","openness":"api","release_date":"2025-08-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.91,"source_primary":"fab-ai.org","source_primary_share":0.5333,"source_primary_group_share":0.5,"capability_breadth":3.0,"capability_primary":"knowledge_general","capability_primary_share":0.5333,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"stablelm-3B-4e1t","provider":"Stability AI","country":"UK","openness":"open","release_date":"2023-09-29","param_count":3.0,"active_param_count":null,"training_compute_flop":6.21e+22,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.04,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.96,"source_primary":"euroeval.com","source_primary_share":0.6,"source_primary_group_share":0.5714,"capability_breadth":3.26,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4667,"capability_primary_group_count":7,"capability_non_primary_group_count":7},{"model":"MiniMax-M1","provider":"MiniMax","country":"CN","openness":"open","release_date":"2025-06-01","param_count":456.0,"active_param_count":45.9,"training_compute_flop":4.3240062e+24,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":14.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":7.0,"source_primary":"raw.githubusercontent.com/llm2014/llm_benchmark","source_primary_share":0.2667,"source_primary_group_share":0.2143,"capability_breadth":4.59,"capability_primary":"reasoning_logical","capability_primary_share":0.4,"capability_primary_group_count":5,"capability_non_primary_group_count":9},{"model":"Intern-S1","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2025-07-27","param_count":241.0,"active_param_count":28.0,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":15,"benchmark_coverage_score":11.39,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.59,"source_primary":"huggingface.co/spaces/internscience/scieval-leaderboard","source_primary_share":0.4667,"source_primary_group_share":0.4667,"capability_breadth":2.85,"capability_primary":"vision_language","capability_primary_share":0.5333,"capability_primary_group_count":8,"capability_non_primary_group_count":7},{"model":"GLM-4.6 (Thinking)","provider":"Zhipu","country":"CN","openness":"open","release_date":"2025-09-30","param_count":357.0,"active_param_count":32.0,"training_compute_flop":4.42e+24,"num_benchmarks":15,"benchmark_group_count":14,"benchmark_coverage_score":13.04,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.28,"source_primary":"raw.githubusercontent.com/llm2014/llm_benchmark","source_primary_share":0.6667,"source_primary_group_share":0.6429,"capability_breadth":3.26,"capability_primary":"code","capability_primary_share":0.4667,"capability_primary_group_count":7,"capability_non_primary_group_count":7},{"model":"DeepSeek-V4-Flash-FP8","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2026-04-23","param_count":284.0,"active_param_count":13.0,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":11,"benchmark_coverage_score":11.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.42,"source_primary":"huggingface.co","source_primary_share":0.6,"source_primary_group_share":0.8182,"capability_breadth":3.0,"capability_primary":"math","capability_primary_share":0.5333,"capability_primary_group_count":4,"capability_non_primary_group_count":7},{"model":"GLM-5.1 FP8","provider":"Zhipu","country":"CN","openness":"open","release_date":"2026-03-27","param_count":744.0,"active_param_count":40.0,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":11,"benchmark_coverage_score":11.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.42,"source_primary":"huggingface.co","source_primary_share":0.6,"source_primary_group_share":0.8182,"capability_breadth":3.0,"capability_primary":"math","capability_primary_share":0.5333,"capability_primary_group_count":4,"capability_non_primary_group_count":7},{"model":"K2-V2-Instruct","provider":"LLM360","country":"UAE","openness":"open","release_date":"2026-01-27","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":15,"benchmark_group_count":11,"benchmark_coverage_score":11.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.42,"source_primary":"huggingface.co","source_primary_share":0.6,"source_primary_group_share":0.8182,"capability_breadth":3.0,"capability_primary":"math","capability_primary_share":0.5333,"capability_primary_group_count":4,"capability_non_primary_group_count":7},{"model":"GLM-4.6 FP8","provider":"Zhipu","country":"CN","openness":"open","release_date":"2025-09-30","param_count":357.0,"active_param_count":32.0,"training_compute_flop":4.42e+24,"num_benchmarks":15,"benchmark_group_count":11,"benchmark_coverage_score":11.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.42,"source_primary":"huggingface.co","source_primary_share":0.6,"source_primary_group_share":0.8182,"capability_breadth":3.0,"capability_primary":"math","capability_primary_share":0.5333,"capability_primary_group_count":4,"capability_non_primary_group_count":7},{"model":"Qwen 3 235B A22B FP8","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":235.0,"active_param_count":22.0,"training_compute_flop":4.752e+24,"num_benchmarks":14,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.52,"source_primary":"huggingface.co/spaces/kluster-ai/llm-hallucination-detection-leaderboard","source_primary_share":0.5,"source_primary_group_share":0.4444,"capability_breadth":5.44,"capability_primary":"knowledge_general","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":7},{"model":"Llama-Krikri-8B-Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2025-02-07","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.08,"source_primary":"huggingface.co/spaces/yangzhang33/greekmmlu-leaderboard","source_primary_share":0.4286,"source_primary_group_share":0.6667,"capability_breadth":3.5,"capability_primary":"knowledge_general","capability_primary_share":0.3571,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Codestral-22B-v0.1","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-05-29","param_count":22.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":14,"benchmark_coverage_score":14.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":4.45,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.4286,"source_primary_group_share":0.4286,"capability_breadth":3.77,"capability_primary":"code","capability_primary_share":0.4286,"capability_primary_group_count":6,"capability_non_primary_group_count":8},{"model":"flan-t5-xxl","provider":"Google","country":"US","openness":"api","release_date":"2022-10-21","param_count":11.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":12,"benchmark_coverage_score":12.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.67,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.4615,"source_primary_group_share":0.5,"capability_breadth":4.83,"capability_primary":"structured_precision","capability_primary_share":0.3077,"capability_primary_group_count":4,"capability_non_primary_group_count":8},{"model":"Orca-2-7B","provider":"Microsoft","country":"US","openness":"open","release_date":"2023-11-14","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.65,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.68,"source_primary":"huggingface.co/spaces/logikon/open_cot_leaderboard","source_primary_share":0.4286,"source_primary_group_share":0.4615,"capability_breadth":3.16,"capability_primary":"reasoning_logical","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"occiglot-7B-es-en-instruct","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-05","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":14,"benchmark_coverage_score":14.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.96,"source_primary":"huggingface.co/spaces/la-leaderboard/la-leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5714,"capability_breadth":3.77,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3571,"capability_primary_group_count":5,"capability_non_primary_group_count":9},{"model":"leniachat-gemma-2B-v0","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-09","param_count":2.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":14,"benchmark_coverage_score":14.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.96,"source_primary":"huggingface.co/spaces/la-leaderboard/la-leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5714,"capability_breadth":3.77,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3571,"capability_primary_group_count":5,"capability_non_primary_group_count":9},{"model":"leniachat-qwen2-1.5B-v0","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-16","param_count":1.5,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":14,"benchmark_coverage_score":14.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.96,"source_primary":"huggingface.co/spaces/la-leaderboard/la-leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5714,"capability_breadth":3.77,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3571,"capability_primary_group_count":5,"capability_non_primary_group_count":9},{"model":"TW3-JRGL-v2","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-01","param_count":72.3,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Rhea-72B-v0.5","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-22","param_count":72.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"MultiVerse_70B","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-25","param_count":72.3,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Le_Triomphant-ECE-TW3","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-01","param_count":72.3,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Smaug-72B-v0.1","provider":"Other","country":"XX","openness":"open","release_date":"2024-02-02","param_count":72.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"UNA-SimpleSmaug-34B-v1beta","provider":"Other","country":"XX","openness":"open","release_date":"2024-02-05","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Smaug-34B-v0.1","provider":"Other","country":"XX","openness":"open","release_date":"2024-01-25","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"ECE-TW3-JRGL-V5","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-10","param_count":72.3,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Yi-34Bx2-MoE-60B-DPO","provider":"01.AI","country":"CN","openness":"open","release_date":"2024-01-23","param_count":60.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Mixtral_34Bx2_MoE_60B","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-01-05","param_count":60.8,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Bagel-Hermes-34B-Slerp","provider":"Jon Durbin","country":"US","openness":"open","release_date":"2024-01-12","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"YiSM-blossom5.1-34B-SLERP","provider":"01.AI","country":"CN","openness":"open","release_date":"2024-08-27","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"bagel-dpo-34B-v0.5","provider":"Jon Durbin","country":"US","openness":"open","release_date":"2024-04-01","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":13.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.63,"capability_primary":"knowledge_general","capability_primary_share":0.3571,"capability_primary_group_count":5,"capability_non_primary_group_count":8},{"model":"YiSM-34B-0rn","provider":"01.AI","country":"CN","openness":"open","release_date":"2024-05-26","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Higgs-Llama-3-70B","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-05","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"blossom-v5.1-34B","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-19","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"openbuddy-yi1.5-34B-v21.3-32k","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-05","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"dolphin-2.9.1-llama-3-70B","provider":"Cognitive Computations","country":"US","openness":"open","release_date":"2024-05-22","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Llama3-TenyxChat-70B","provider":"Meta","country":"US","openness":"open","release_date":"2024-04-26","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"llama-3-70B-Instruct-abliterated","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-07","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Meta-LLama-3-Cat-Smaug-LLama-70B","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-24","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Llama 3 70B Japanese Suzume Vector V0.1","provider":"Meta","country":"US","openness":"open","release_date":"2024-04-28","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"dolphin-2.9.1-yi-1.5-34B","provider":"Cognitive Computations","country":"US","openness":"open","release_date":"2024-05-18","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"34B-beta","provider":"Other","country":"XX","openness":"open","release_date":"2024-02-06","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"autotrain-llama3-70B-orpo-v1","provider":"Other","country":"XX","openness":"open","release_date":null,"param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"openbuddy-zero-56B-v21.2-32k","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-10","param_count":56.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Smaug-Llama-3-70B-Instruct-32K","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-11","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Llama 3 70B ShiningValiant2","provider":"Meta","country":"US","openness":"open","release_date":"2024-04-20","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"openbuddy-llama3-70B-v21.2-32k","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-12","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"dolphin-2.9.3-Yi-1.5-34B-32k","provider":"Cognitive Computations","country":"US","openness":"open","release_date":"2024-06-23","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"ECE-TW3-JRGL-V1","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-03","param_count":69.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"openbuddy-yi1.5-9B-v21.1-32k","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-22","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":12,"benchmark_coverage_score":10.59,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.95,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5833,"capability_breadth":2.65,"capability_primary":"knowledge_general","capability_primary_share":0.4286,"capability_primary_group_count":5,"capability_non_primary_group_count":7},{"model":"GritLM-8x7B-KTO","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-17","param_count":56.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"LogoS-7Bx2-MoE-13B-v0.2","provider":"Other","country":"XX","openness":"open","release_date":"2024-01-21","param_count":13.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Midnight-Miqu-70B-v1.5","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-11","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"NeuralSynthesis-7B-v0.3","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-07","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"openbuddy-zero-14B-v22.3-32k","provider":"Other","country":"XX","openness":"open","release_date":"2024-07-16","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Mixolar-4x7B","provider":"Other","country":"XX","openness":"open","release_date":"2023-12-30","param_count":28.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"CognitiveFusion2-4x7B-BF16","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-06","param_count":28.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"YamshadowExperiment28-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-18","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Meta-Llama-3-70B-Instruct-abliterated-v3.5","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-28","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"MixtureofMerges-MoE-4x7B-v5","provider":"Other","country":"XX","openness":"open","release_date":"2024-02-25","param_count":28.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"mera-mix-4x7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-13","param_count":28.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"multimaster-7B-v6","provider":"Other","country":"XX","openness":"open","release_date":"2024-02-24","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Smaug-Mixtral-v0.1","provider":"Other","country":"XX","openness":"open","release_date":"2024-02-18","param_count":46.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"MultiverseEx26-7B-slerp","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-31","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"MixtureofMerges-MoE-4x7B-v4","provider":"Other","country":"XX","openness":"open","release_date":"2024-02-11","param_count":28.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"NeuralExperiment-7B-MagicCoder-v7.5","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-07","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Llama 3 70B Fireplace","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-09","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Mengzi3-8B-Chat","provider":"Langboat","country":"CN","openness":"open","release_date":"2024-09-14","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"ChimeraLlama-3-8B-v2","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-22","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"openbuddy-llama3-8B-v21.2-32k","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-18","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Einstein-v6.1-Llama3-8B","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-19","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"janus-dpo-7B","provider":"KAIST","country":"KR","openness":"open","release_date":"2024-04-25","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"janus-7B","provider":"KAIST","country":"KR","openness":"open","release_date":"2024-04-04","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"CodeBooga-34B-v0.1","provider":"Other","country":"XX","openness":"open","release_date":"2023-10-19","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"speechless-zephyr-code-functionary-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-01-23","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"ALMA-13B-R","provider":"Other","country":"XX","openness":"open","release_date":"2024-01-17","param_count":13.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Llama-3-Refueled","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-03","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Infinirc-Llama3-8B-2G-Release-v1.0","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-26","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"blossom-v5.1-9B","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-15","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"L-MChat-Small","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-11","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":12.25,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.5714,"source_primary_group_share":0.5385,"capability_breadth":3.06,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Llama 3.1 8B Ita","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-23","param_count":8.0,"active_param_count":null,"training_compute_flop":1.224e+24,"num_benchmarks":14,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.98,"source_primary":"huggingface.co/spaces/mii-llm/pinocchio-ita-leaderboard","source_primary_share":0.5,"source_primary_group_share":0.6667,"capability_breadth":4.26,"capability_primary":"knowledge_multilingual","capability_primary_share":0.2857,"capability_primary_group_count":3,"capability_non_primary_group_count":6},{"model":"Qwen2-1.5B-Ita","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-06-13","param_count":1.5,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.98,"source_primary":"huggingface.co/spaces/mii-llm/pinocchio-ita-leaderboard","source_primary_share":0.5,"source_primary_group_share":0.6667,"capability_breadth":4.26,"capability_primary":"knowledge_multilingual","capability_primary_share":0.2857,"capability_primary_group_count":3,"capability_non_primary_group_count":6},{"model":"Hermes-2-Pro-Llama-3-8B","provider":"Nous Research","country":"US","openness":"open","release_date":"2024-04-30","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.5,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.4615,"source_primary_group_share":0.6,"capability_breadth":3.6,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4615,"capability_primary_group_count":3,"capability_non_primary_group_count":7},{"model":"Qwen 3 Coder Plus","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-09-23","param_count":480.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":14,"benchmark_coverage_score":14.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":4.45,"source_primary":"benchmarklist.com","source_primary_share":0.2857,"source_primary_group_share":0.2857,"capability_breadth":3.5,"capability_primary":"code","capability_primary_share":0.5,"capability_primary_group_count":7,"capability_non_primary_group_count":7},{"model":"Mistral-Large-3-675B-Instruct-2512","provider":"Mistral","country":"FR","openness":"open","release_date":"2025-12-02","param_count":675.0,"active_param_count":41.0,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.94,"source_primary":"github.com/geralt-targaryen/mc-evaluation","source_primary_share":0.4286,"source_primary_group_share":0.5,"capability_breadth":3.06,"capability_primary":"math","capability_primary_share":0.5,"capability_primary_group_count":3,"capability_non_primary_group_count":7},{"model":"Grok 4.20 Multi-Agent","provider":"xAI","country":"US","openness":"api","release_date":"2026-03-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":12,"benchmark_coverage_score":12.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":7.2,"source_primary":"arena.ai","source_primary_share":0.2308,"source_primary_group_share":0.25,"capability_breadth":3.76,"capability_primary":"knowledge_general","capability_primary_share":0.3846,"capability_primary_group_count":5,"capability_non_primary_group_count":7},{"model":"GPT-4o (Mar 2025)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-03-01","param_count":200.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":14,"benchmark_coverage_score":14.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":2.8,"source_primary":"artificialanalysis.ai","source_primary_share":0.5714,"source_primary_group_share":0.5714,"capability_breadth":4.9,"capability_primary":"knowledge_general","capability_primary_share":0.2857,"capability_primary_group_count":4,"capability_non_primary_group_count":10},{"model":"Llama3.1-Aloe-Beta-8B","provider":"Meta","country":"US","openness":"open","release_date":"2024-10-30","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":10,"benchmark_coverage_score":9.52,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.92,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.6,"capability_breadth":2.38,"capability_primary":"knowledge_general","capability_primary_share":0.6,"capability_primary_group_count":6,"capability_non_primary_group_count":4},{"model":"GPT-5.1 (Thinking)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-11-12","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":12,"benchmark_coverage_score":12.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.12,"source_primary":"scale.com","source_primary_share":0.6923,"source_primary_group_share":0.6667,"capability_breadth":3.76,"capability_primary":"knowledge_general","capability_primary_share":0.4615,"capability_primary_group_count":5,"capability_non_primary_group_count":7},{"model":"Gemini 3.1 Pro (Preview) (Thinking)","provider":"Google","country":"US","openness":"api","release_date":"2026-02-19","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":14,"benchmark_group_count":13,"benchmark_coverage_score":11.88,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.96,"source_primary":"balrogai.com","source_primary_share":0.4286,"source_primary_group_share":0.4615,"capability_breadth":2.97,"capability_primary":"reasoning_logical","capability_primary_share":0.4286,"capability_primary_group_count":5,"capability_non_primary_group_count":8},{"model":"Qwen 3 Max (2026-01-23)","provider":"Alibaba","country":"CN","openness":"api","release_date":"2026-01-23","param_count":700.0,"active_param_count":null,"training_compute_flop":1.512e+25,"num_benchmarks":13,"benchmark_group_count":13,"benchmark_coverage_score":13.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.25,"source_primary":"vals.ai","source_primary_share":0.6154,"source_primary_group_share":0.6154,"capability_breadth":3.31,"capability_primary":"knowledge_general","capability_primary_share":0.4615,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"plutus-8B-instruct","provider":"Other","country":"XX","openness":"open","release_date":"2025-02-18","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":13,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.29,"source_primary":"huggingface.co/spaces/yangzhang33/greekmmlu-leaderboard","source_primary_share":0.4615,"source_primary_group_share":0.625,"capability_breadth":2.68,"capability_primary":"knowledge_general","capability_primary_share":0.4615,"capability_primary_group_count":5,"capability_non_primary_group_count":3},{"model":"OpenCoder-1.5B-Instruct","provider":"Other","country":"XX","openness":"open","release_date":"2024-11-07","param_count":1.5,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":13,"benchmark_group_count":11,"benchmark_coverage_score":11.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.81,"source_primary":"euroeval.com","source_primary_share":0.7692,"source_primary_group_share":0.7273,"capability_breadth":3.07,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4615,"capability_primary_group_count":6,"capability_non_primary_group_count":5},{"model":"Claude Instant 1.2","provider":"Anthropic","country":"US","openness":"api","release_date":"2023-08-09","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 1/2 family retired or superseded by newer Claude models.","num_benchmarks":13,"benchmark_group_count":12,"benchmark_coverage_score":11.86,"source_family_count":5,"source_group_family_count":4,"source_effective_breadth":2.77,"source_primary":"epoch.ai","source_primary_share":0.3846,"source_primary_group_share":0.4167,"capability_breadth":2.96,"capability_primary":"knowledge_general","capability_primary_share":0.5385,"capability_primary_group_count":6,"capability_non_primary_group_count":6},{"model":"Orca-2-13B","provider":"Microsoft","country":"US","openness":"open","release_date":"2023-11-14","param_count":13.0,"active_param_count":null,"training_compute_flop":4.6e+22,"num_benchmarks":13,"benchmark_group_count":12,"benchmark_coverage_score":11.08,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.32,"source_primary":"huggingface.co/spaces/logikon/open_cot_leaderboard","source_primary_share":0.4615,"source_primary_group_share":0.5,"capability_breadth":2.77,"capability_primary":"reasoning_logical","capability_primary_share":0.5385,"capability_primary_group_count":6,"capability_non_primary_group_count":6},{"model":"internlm2-20B","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2024-01-12","param_count":20.0,"active_param_count":null,"training_compute_flop":3.12e+23,"num_benchmarks":13,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.59,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.6154,"source_primary_group_share":0.7778,"capability_breadth":2.68,"capability_primary":"knowledge_general","capability_primary_share":0.4615,"capability_primary_group_count":5,"capability_non_primary_group_count":4},{"model":"Lexora-Lite-3B","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-19","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":13,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/mii-llm/pinocchio-ita-leaderboard","source_primary_share":0.5385,"source_primary_group_share":0.75,"capability_breadth":4.33,"capability_primary":"knowledge_general","capability_primary_share":0.3077,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"DeciLM-7B","provider":"Deci AI","country":"IL","openness":"open","release_date":"2023-12-10","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":13,"benchmark_group_count":12,"benchmark_coverage_score":10.09,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.32,"source_primary":"huggingface.co/spaces/logikon/open_cot_leaderboard","source_primary_share":0.4615,"source_primary_group_share":0.5,"capability_breadth":2.52,"capability_primary":"reasoning_logical","capability_primary_share":0.5385,"capability_primary_group_count":6,"capability_non_primary_group_count":6},{"model":"SuperNova-Medius","provider":"Arcee AI","country":"US","openness":"open","release_date":"2024-10-02","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":13,"benchmark_group_count":12,"benchmark_coverage_score":11.08,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.32,"source_primary":"huggingface.co/spaces/logikon/open_cot_leaderboard","source_primary_share":0.4615,"source_primary_group_share":0.5,"capability_breadth":2.77,"capability_primary":"reasoning_logical","capability_primary_share":0.5385,"capability_primary_group_count":6,"capability_non_primary_group_count":6},{"model":"LLaMA3-iterative-DPO-final","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-17","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":13,"benchmark_group_count":13,"benchmark_coverage_score":13.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.4615,"source_primary_group_share":0.4615,"capability_breadth":3.6,"capability_primary":"reasoning_logical","capability_primary_share":0.4615,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"openbuddy-llama3-8B-v21.1-8k","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-20","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":13,"benchmark_group_count":12,"benchmark_coverage_score":11.08,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.32,"source_primary":"huggingface.co/spaces/logikon/open_cot_leaderboard","source_primary_share":0.4615,"source_primary_group_share":0.5,"capability_breadth":2.77,"capability_primary":"reasoning_logical","capability_primary_share":0.5385,"capability_primary_group_count":6,"capability_non_primary_group_count":6},{"model":"Qwen 2.5 Coder 14B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-11-12","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":13,"benchmark_group_count":12,"benchmark_coverage_score":12.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.5,"source_primary_group_share":0.5,"capability_breadth":3.79,"capability_primary":"knowledge_general","capability_primary_share":0.3333,"capability_primary_group_count":4,"capability_non_primary_group_count":8},{"model":"neural-chat-7B-v3-1","provider":"Other","country":"XX","openness":"open","release_date":"2023-11-14","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":13,"benchmark_group_count":12,"benchmark_coverage_score":11.08,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.32,"source_primary":"huggingface.co/spaces/logikon/open_cot_leaderboard","source_primary_share":0.4615,"source_primary_group_share":0.5,"capability_breadth":2.77,"capability_primary":"reasoning_logical","capability_primary_share":0.5385,"capability_primary_group_count":6,"capability_non_primary_group_count":6},{"model":"Gemini 3.1 Flash","provider":"Google","country":"US","openness":"api","release_date":"2026-03-03","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":13,"benchmark_group_count":13,"benchmark_coverage_score":12.29,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.61,"source_primary":"benchmarklist.com","source_primary_share":0.7692,"source_primary_group_share":0.7692,"capability_breadth":3.07,"capability_primary":"vision_language","capability_primary_share":0.4615,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Claude Opus 4.8 (xHigh)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-05-29","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":13,"benchmark_group_count":12,"benchmark_coverage_score":11.08,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":8.0,"source_primary":"raw.githubusercontent.com/llm2014/llm_benchmark","source_primary_share":0.2308,"source_primary_group_share":0.1667,"capability_breadth":2.77,"capability_primary":"code","capability_primary_share":0.5385,"capability_primary_group_count":7,"capability_non_primary_group_count":5},{"model":"Seed 2.0 Lite","provider":"ByteDance","country":"CN","openness":"api","release_date":"2026-02-16","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":13,"benchmark_group_count":11,"benchmark_coverage_score":11.0,"source_family_count":11,"source_group_family_count":9,"source_effective_breadth":8.07,"source_primary":"dubesor.de","source_primary_share":0.1538,"source_primary_group_share":0.1818,"capability_breadth":5.45,"capability_primary":"agentic","capability_primary_share":0.3077,"capability_primary_group_count":3,"capability_non_primary_group_count":8},{"model":"O3 (2025-04-16) (Medium)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-04-16","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":13,"benchmark_group_count":12,"benchmark_coverage_score":12.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":5.54,"source_primary":"scale.com","source_primary_share":0.25,"source_primary_group_share":0.25,"capability_breadth":5.14,"capability_primary":"knowledge_general","capability_primary_share":0.3333,"capability_primary_group_count":4,"capability_non_primary_group_count":8},{"model":"LongCat-Flash-Thinking-2601","provider":"Meituan","country":"CN","openness":"api","release_date":"2026-01-29","param_count":560.0,"active_param_count":27.0,"training_compute_flop":null,"num_benchmarks":13,"benchmark_group_count":12,"benchmark_coverage_score":12.0,"source_family_count":8,"source_group_family_count":7,"source_effective_breadth":5.54,"source_primary":"llm-stats.com","source_primary_share":0.2308,"source_primary_group_share":0.25,"capability_breadth":3.76,"capability_primary":"agentic","capability_primary_share":0.3846,"capability_primary_group_count":4,"capability_non_primary_group_count":8},{"model":"GPT-5 Nano (Low)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-08-07","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":13,"benchmark_group_count":13,"benchmark_coverage_score":13.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":6.26,"source_primary":"huggingface.co/spaces/aiml-tuda/slr-leaderboard","source_primary_share":0.3077,"source_primary_group_share":0.3077,"capability_breadth":4.57,"capability_primary":"reasoning_logical","capability_primary_share":0.3846,"capability_primary_group_count":5,"capability_non_primary_group_count":8},{"model":"Llama3","provider":"Meta","country":"US","openness":"open","release_date":"2024-04-18","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":13,"benchmark_group_count":12,"benchmark_coverage_score":9.52,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.32,"source_primary":"drchangliu.github.io","source_primary_share":0.4615,"source_primary_group_share":0.5,"capability_breadth":2.38,"capability_primary":"reasoning_logical","capability_primary_share":0.6154,"capability_primary_group_count":7,"capability_non_primary_group_count":5},{"model":"RedPajama-INCITE-7B-Base","provider":"Together","country":"US","openness":"open","release_date":"2023-05-05","param_count":7.0,"active_param_count":null,"training_compute_flop":4.1e+22,"num_benchmarks":13,"benchmark_group_count":13,"benchmark_coverage_score":12.29,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"epoch.ai","source_primary_share":0.5385,"source_primary_group_share":0.5385,"capability_breadth":3.07,"capability_primary":"knowledge_general","capability_primary_share":0.4615,"capability_primary_group_count":6,"capability_non_primary_group_count":7},{"model":"Llama-3.1-Tulu-3-8B-RM","provider":"Meta","country":"US","openness":"open","release_date":"2024-11-20","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":13,"benchmark_group_count":13,"benchmark_coverage_score":13.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.99,"source_primary":"huggingface.co","source_primary_share":0.5385,"source_primary_group_share":0.5385,"capability_breadth":4.33,"capability_primary":"instruction_following","capability_primary_share":0.3077,"capability_primary_group_count":4,"capability_non_primary_group_count":9},{"model":"INTELLECT-1-Instruct","provider":"PrimeIntellect","country":"US","openness":"open","release_date":"2024-09-01","param_count":10.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":13,"benchmark_group_count":12,"benchmark_coverage_score":12.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":2.0,"source_primary":"epoch.ai","source_primary_share":0.5385,"source_primary_group_share":0.5,"capability_breadth":3.6,"capability_primary":"reasoning_logical","capability_primary_share":0.3846,"capability_primary_group_count":5,"capability_non_primary_group_count":7},{"model":"GLM-4.5 (Thinking)","provider":"Zhipu","country":"CN","openness":"open","release_date":"2025-07-28","param_count":355.0,"active_param_count":32.0,"training_compute_flop":4.42e+24,"num_benchmarks":13,"benchmark_group_count":12,"benchmark_coverage_score":7.43,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.12,"source_primary":"raw.githubusercontent.com/llm2014/llm_benchmark","source_primary_share":0.6923,"source_primary_group_share":0.6667,"capability_breadth":1.86,"capability_primary":"code","capability_primary_share":0.6923,"capability_primary_group_count":9,"capability_non_primary_group_count":3},{"model":"Grok 4.20 (Non-reasoning)","provider":"xAI","country":"US","openness":"api","release_date":"2026-03-10","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":13,"benchmark_group_count":11,"benchmark_coverage_score":11.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":4.17,"source_primary":"designarena.ai","source_primary_share":0.4615,"source_primary_group_share":0.3636,"capability_breadth":8.05,"capability_primary":"reasoning_logical","capability_primary_share":0.1538,"capability_primary_group_count":2,"capability_non_primary_group_count":9},{"model":"ERNIE-4.5-21B-A3B (Thinking)","provider":"Baidu","country":"CN","openness":"open","release_date":"2025-06-28","param_count":21.0,"active_param_count":3.0,"training_compute_flop":1.8e+23,"num_benchmarks":12,"benchmark_group_count":11,"benchmark_coverage_score":11.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":2.95,"source_primary":"rank.opencompass.org.cn","source_primary_share":0.5833,"source_primary_group_share":0.5455,"capability_breadth":4.0,"capability_primary":"knowledge_general","capability_primary_share":0.4167,"capability_primary_group_count":5,"capability_non_primary_group_count":6},{"model":"GPT-5.5 Pro (xHigh)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-04-23","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":12,"benchmark_coverage_score":10.29,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.67,"source_primary":"epoch.ai","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":2.57,"capability_primary":"math","capability_primary_share":0.5833,"capability_primary_group_count":7,"capability_non_primary_group_count":5},{"model":"deepseek-llm-67B-chat","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2023-11-29","param_count":67.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":12,"benchmark_coverage_score":12.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.43,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.5,"source_primary_group_share":0.5,"capability_breadth":6.0,"capability_primary":"knowledge_general","capability_primary_share":0.25,"capability_primary_group_count":3,"capability_non_primary_group_count":9},{"model":"Mistral-7B-v0.2","provider":"Mistral","country":"FR","openness":"open","release_date":"2023-12-06","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":12,"benchmark_coverage_score":12.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":3.43,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.5,"source_primary_group_share":0.5,"capability_breadth":5.54,"capability_primary":"reasoning_logical","capability_primary_share":0.3333,"capability_primary_group_count":4,"capability_non_primary_group_count":8},{"model":"Skywork-o1-Open-Llama-3.1-8B","provider":"Skywork","country":"CN","openness":"open","release_date":"2024-11-26","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.5,"source_primary_group_share":0.8571,"capability_breadth":3.6,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4167,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"internlm2-base-7B","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2024-01-12","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.875,"capability_breadth":2.77,"capability_primary":"knowledge_general","capability_primary_share":0.4167,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"Nemotron 3 Super","provider":"NVIDIA","country":"US","openness":"open","release_date":"2026-03-11","param_count":120.0,"active_param_count":12.0,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":11,"benchmark_coverage_score":11.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.46,"source_primary":"benchmarklist.com","source_primary_share":0.8182,"source_primary_group_share":0.8182,"capability_breadth":6.37,"capability_primary":"code","capability_primary_share":0.1818,"capability_primary_group_count":2,"capability_non_primary_group_count":9},{"model":"Darkest-muse-v1","provider":"Other","country":"XX","openness":"open","release_date":"2024-10-22","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":11,"benchmark_coverage_score":11.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.57,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.5455,"source_primary_group_share":0.5455,"capability_breadth":3.9,"capability_primary":"knowledge_general","capability_primary_share":0.3636,"capability_primary_group_count":4,"capability_non_primary_group_count":7},{"model":"gemma-2-Ifable-9B","provider":"Google","country":"US","openness":"api","release_date":"2024-09-10","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":11,"benchmark_coverage_score":11.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.57,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.5455,"source_primary_group_share":0.5455,"capability_breadth":3.9,"capability_primary":"knowledge_general","capability_primary_share":0.3636,"capability_primary_group_count":4,"capability_non_primary_group_count":7},{"model":"GPT-5.3 Codex (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-02-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":12,"benchmark_coverage_score":12.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":5.14,"source_primary":"mercor.com","source_primary_share":0.3333,"source_primary_group_share":0.3333,"capability_breadth":3.79,"capability_primary":"knowledge_general","capability_primary_share":0.4167,"capability_primary_group_count":5,"capability_non_primary_group_count":7},{"model":"Palmyra Fin","provider":"Writer","country":"US","openness":"api","release_date":"2024-07-30","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":12,"benchmark_coverage_score":8.73,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.32,"source_primary":"storage.googleapis.com","source_primary_share":0.5,"source_primary_group_share":0.5,"capability_breadth":2.18,"capability_primary":"instruction_following","capability_primary_share":0.5833,"capability_primary_group_count":7,"capability_non_primary_group_count":5},{"model":"Qwen3 235B A22B FP8 Throughput","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-04-28","param_count":235.0,"active_param_count":22.0,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":12,"benchmark_coverage_score":8.73,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.32,"source_primary":"storage.googleapis.com","source_primary_share":0.5,"source_primary_group_share":0.5,"capability_breadth":2.18,"capability_primary":"instruction_following","capability_primary_share":0.5833,"capability_primary_group_count":7,"capability_non_primary_group_count":5},{"model":"Palmyra Med","provider":"Writer","country":"US","openness":"api","release_date":"2024-07-30","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":12,"benchmark_coverage_score":8.73,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.32,"source_primary":"storage.googleapis.com","source_primary_share":0.5,"source_primary_group_share":0.5,"capability_breadth":2.18,"capability_primary":"instruction_following","capability_primary_share":0.5833,"capability_primary_group_count":7,"capability_non_primary_group_count":5},{"model":"Llama-3.1-Argunaut-1-8B-SFT","provider":"Meta","country":"US","openness":"open","release_date":"2024-12-31","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":11,"benchmark_coverage_score":9.6,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.98,"source_primary":"huggingface.co/spaces/logikon/open_cot_leaderboard","source_primary_share":0.5,"source_primary_group_share":0.5455,"capability_breadth":2.4,"capability_primary":"reasoning_logical","capability_primary_share":0.5833,"capability_primary_group_count":6,"capability_non_primary_group_count":5},{"model":"DeciLM-7B-instruct","provider":"Deci AI","country":"IL","openness":"open","release_date":"2023-12-10","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":11,"benchmark_coverage_score":9.6,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.98,"source_primary":"huggingface.co/spaces/logikon/open_cot_leaderboard","source_primary_share":0.5,"source_primary_group_share":0.5455,"capability_breadth":2.4,"capability_primary":"reasoning_logical","capability_primary_share":0.5833,"capability_primary_group_count":6,"capability_non_primary_group_count":5},{"model":"stablelm-2-12B-chat","provider":"Stability AI","country":"UK","openness":"open","release_date":"2024-04-04","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":12,"benchmark_coverage_score":12.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":2.0,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.5,"source_primary_group_share":0.5,"capability_breadth":3.13,"capability_primary":"reasoning_logical","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":6},{"model":"zephyr-7B-gemma-v0.1","provider":"HuggingFace","country":"FR","openness":"open","release_date":"2024-03-01","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":12,"benchmark_coverage_score":12.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":2.0,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.5,"source_primary_group_share":0.5,"capability_breadth":3.13,"capability_primary":"reasoning_logical","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":6},{"model":"Matter-0.2-7B-DPO","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-13","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":11,"benchmark_coverage_score":9.6,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.98,"source_primary":"huggingface.co/spaces/logikon/open_cot_leaderboard","source_primary_share":0.5,"source_primary_group_share":0.5455,"capability_breadth":2.4,"capability_primary":"reasoning_logical","capability_primary_share":0.5833,"capability_primary_group_count":6,"capability_non_primary_group_count":5},{"model":"stablelm-2-zephyr-1.6B","provider":"Stability AI","country":"UK","openness":"open","release_date":"2024-01-19","param_count":1.6,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":12,"benchmark_coverage_score":12.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":2.0,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.5,"source_primary_group_share":0.5,"capability_breadth":3.13,"capability_primary":"reasoning_logical","capability_primary_share":0.5,"capability_primary_group_count":6,"capability_non_primary_group_count":6},{"model":"merlinite-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-02","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":11,"benchmark_coverage_score":9.6,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.98,"source_primary":"huggingface.co/spaces/logikon/open_cot_leaderboard","source_primary_share":0.5,"source_primary_group_share":0.5455,"capability_breadth":2.4,"capability_primary":"reasoning_logical","capability_primary_share":0.5833,"capability_primary_group_count":6,"capability_non_primary_group_count":5},{"model":"Claude Opus 4 (Thinking 16K)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-05-22","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 4.0 family deprecated in favor of newer Claude 4.x models.","num_benchmarks":12,"benchmark_group_count":11,"benchmark_coverage_score":11.0,"source_family_count":9,"source_group_family_count":9,"source_effective_breadth":8.07,"source_primary":"arcprize.org","source_primary_share":0.1667,"source_primary_group_share":0.1818,"capability_breadth":4.5,"capability_primary":"reasoning_logical","capability_primary_share":0.3333,"capability_primary_group_count":4,"capability_non_primary_group_count":7},{"model":"Qwen Max","provider":"Alibaba","country":"CN","openness":"api","release_date":"2025-04-28","param_count":700.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":5.4,"source_primary":"arxiv.org","source_primary_share":0.4545,"source_primary_group_share":0.3333,"capability_breadth":5.26,"capability_primary":"structured_precision","capability_primary_share":0.2727,"capability_primary_group_count":2,"capability_non_primary_group_count":7},{"model":"Claude 3.7 Sonnet (High)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-02-24","param_count":null,"active_param_count":null,"training_compute_flop":3.35e+25,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 3 family retired or superseded by newer Claude models.","num_benchmarks":12,"benchmark_group_count":10,"benchmark_coverage_score":7.2,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.22,"source_primary":"hal.cs.princeton.edu","source_primary_share":0.9167,"source_primary_group_share":0.9,"capability_breadth":1.8,"capability_primary":"agentic","capability_primary_share":0.6667,"capability_primary_group_count":6,"capability_non_primary_group_count":4},{"model":"internlm2.5-7B","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2024-06-28","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.25,"source_primary":"huggingface.co/spaces/gsma/open-telco-llm-leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.8889,"capability_breadth":3.27,"capability_primary":"knowledge_general","capability_primary_share":0.4167,"capability_primary_group_count":5,"capability_non_primary_group_count":4},{"model":"NVIDIA-Nemotron-3-Super-120B-A12B-FP8","provider":"NVIDIA","country":"US","openness":"open","release_date":"2025-12-15","param_count":120.0,"active_param_count":12.0,"training_compute_flop":null,"num_benchmarks":12,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.13,"source_primary":"github.com/geralt-targaryen/mc-evaluation","source_primary_share":0.5,"source_primary_group_share":0.625,"capability_breadth":2.57,"capability_primary":"math","capability_primary_share":0.5833,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"O3 (Low)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-04-16","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":5.0,"source_primary":"huggingface.co/spaces/dontplantoend/ugi-leaderboard","source_primary_share":0.3,"source_primary_group_share":0.3,"capability_breadth":5.0,"capability_primary":"knowledge_general","capability_primary_share":0.3,"capability_primary_group_count":3,"capability_non_primary_group_count":7},{"model":"CodeQwen1.5-7B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-04-16","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":5,"source_group_family_count":4,"source_effective_breadth":2.58,"source_primary":"epoch.ai","source_primary_share":0.3636,"source_primary_group_share":0.5714,"capability_breadth":2.2,"capability_primary":"code","capability_primary_share":0.6364,"capability_primary_group_count":3,"capability_non_primary_group_count":4},{"model":"deepseek-coder-6.7B-base","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2023-11-02","param_count":6.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.79,"source_primary":"epoch.ai","source_primary_share":0.5556,"source_primary_group_share":0.5556,"capability_breadth":3.52,"capability_primary":"code","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"deepseek-coder-33B-base","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2023-10-28","param_count":33.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.79,"source_primary":"epoch.ai","source_primary_share":0.5556,"source_primary_group_share":0.5556,"capability_breadth":3.52,"capability_primary":"code","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"mistral-tiny","provider":"Mistral","country":"FR","openness":"open","release_date":"2023-12-04","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.59,"source_primary":"lk.instruct.kr","source_primary_share":0.8182,"source_primary_group_share":0.7778,"capability_breadth":2.57,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5455,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"internlm-7B","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2023-07-06","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":11,"benchmark_coverage_score":11.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.81,"source_primary":"epoch.ai","source_primary_share":0.7273,"source_primary_group_share":0.7273,"capability_breadth":5.76,"capability_primary":"knowledge_general","capability_primary_share":0.2727,"capability_primary_group_count":3,"capability_non_primary_group_count":8},{"model":"Qwen 2.5 Turbo","provider":"Alibaba","country":"CN","openness":"api","release_date":"2024-09-19","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":11,"benchmark_coverage_score":11.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.05,"source_primary":"artificialanalysis.ai","source_primary_share":0.6364,"source_primary_group_share":0.6364,"capability_breadth":4.17,"capability_primary":"code","capability_primary_share":0.3636,"capability_primary_group_count":4,"capability_non_primary_group_count":7},{"model":"O1 (2024-12-17) (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2024-12-17","param_count":200.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI o1 family deprecated or superseded by newer o-series models.","num_benchmarks":11,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":4.26,"source_primary":"github.com/lmarena/arena-hard-auto","source_primary_share":0.3,"source_primary_group_share":0.3333,"capability_breadth":3.57,"capability_primary":"creativity","capability_primary_share":0.4,"capability_primary_group_count":3,"capability_non_primary_group_count":6},{"model":"Yi Large (Preview)","provider":"01.AI","country":"CN","openness":"api","release_date":"2024-06-17","param_count":null,"active_param_count":null,"training_compute_flop":1.8e+24,"num_benchmarks":11,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":7,"source_group_family_count":6,"source_effective_breadth":3.33,"source_primary":"crfm.stanford.edu","source_primary_share":0.4545,"source_primary_group_share":0.5,"capability_breadth":3.67,"capability_primary":"knowledge_general","capability_primary_share":0.4545,"capability_primary_group_count":4,"capability_non_primary_group_count":6},{"model":"flan-t5-xl","provider":"Google","country":"US","openness":"api","release_date":"2022-10-21","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.92,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.6,"capability_breadth":3.85,"capability_primary":"structured_precision","capability_primary_share":0.4,"capability_primary_group_count":4,"capability_non_primary_group_count":6},{"model":"flan-t5-large","provider":"Google","country":"US","openness":"api","release_date":"2022-10-21","param_count":0.78,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.92,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.6,"capability_breadth":3.85,"capability_primary":"structured_precision","capability_primary_share":0.4,"capability_primary_group_count":4,"capability_non_primary_group_count":6},{"model":"flan-t5-base","provider":"Google","country":"US","openness":"api","release_date":"2022-10-21","param_count":0.25,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.92,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.6,"capability_breadth":3.85,"capability_primary":"structured_precision","capability_primary_share":0.4,"capability_primary_group_count":4,"capability_non_primary_group_count":6},{"model":"flan-t5-small","provider":"Google","country":"US","openness":"api","release_date":"2022-10-21","param_count":0.08,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.92,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.6,"capability_breadth":3.85,"capability_primary":"structured_precision","capability_primary_share":0.4,"capability_primary_group_count":4,"capability_non_primary_group_count":6},{"model":"Ministrations-8B-v1","provider":"Other","country":"XX","openness":"open","release_date":"2024-11-07","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.17,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.6,"capability_breadth":3.85,"capability_primary":"knowledge_general","capability_primary_share":0.4,"capability_primary_group_count":4,"capability_non_primary_group_count":6},{"model":"mamba-2.8B-hf","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-05","param_count":2.8,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":9,"benchmark_coverage_score":7.22,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.25,"source_primary":"euroeval.com","source_primary_share":0.9091,"source_primary_group_share":0.8889,"capability_breadth":1.81,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7273,"capability_primary_group_count":7,"capability_non_primary_group_count":2},{"model":"DeepSeek V2","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2024-05-06","param_count":236.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.85,"source_primary":"epoch.ai","source_primary_share":0.7273,"source_primary_group_share":0.7,"capability_breadth":2.81,"capability_primary":"reasoning_logical","capability_primary_share":0.4545,"capability_primary_group_count":5,"capability_non_primary_group_count":5},{"model":"Yi-Coder-9B","provider":"01.AI","country":"CN","openness":"open","release_date":"2024-08-15","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":1.92,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.7273,"source_primary_group_share":0.7,"capability_breadth":3.46,"capability_primary":"knowledge_general","capability_primary_share":0.4545,"capability_primary_group_count":4,"capability_non_primary_group_count":6},{"model":"granite-8B-code-base","provider":"IBM","country":"US","openness":"open","release_date":"2024-04-21","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.59,"source_primary":"euroeval.com","source_primary_share":0.8182,"source_primary_group_share":0.7778,"capability_breadth":2.69,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5455,"capability_primary_group_count":5,"capability_non_primary_group_count":4},{"model":"Gemmasutra-9B-v1","provider":"Other","country":"XX","openness":"open","release_date":"2024-07-17","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.17,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.6,"capability_breadth":4.17,"capability_primary":"knowledge_general","capability_primary_share":0.4,"capability_primary_group_count":4,"capability_non_primary_group_count":6},{"model":"Mistral-Nemo-Instruct-2407-abliterated","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-08-15","param_count":12.2,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.17,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.6,"capability_breadth":4.17,"capability_primary":"knowledge_general","capability_primary_share":0.4,"capability_primary_group_count":4,"capability_non_primary_group_count":6},{"model":"Llama-3.1-8B-Lexi-Uncensored-V2","provider":"Meta","country":"US","openness":"open","release_date":"2024-08-09","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.17,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.6,"capability_breadth":4.17,"capability_primary":"knowledge_general","capability_primary_share":0.4,"capability_primary_group_count":4,"capability_non_primary_group_count":6},{"model":"Configurable-Llama-3.1-8B-Instruct","provider":"Other","country":"XX","openness":"open","release_date":"2024-07-24","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.17,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.6,"capability_breadth":4.17,"capability_primary":"knowledge_general","capability_primary_share":0.4,"capability_primary_group_count":4,"capability_non_primary_group_count":6},{"model":"Nanbeige4-3B-Thinking-2511","provider":"Nanbeige","country":"CN","openness":"open","release_date":"2025-11-01","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":10,"benchmark_coverage_score":9.52,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":3.57,"source_primary":"benchmarklist.com","source_primary_share":0.3,"source_primary_group_share":0.3,"capability_breadth":2.38,"capability_primary":"creativity","capability_primary_share":0.6,"capability_primary_group_count":6,"capability_non_primary_group_count":4},{"model":"OLMo 2 32B Instruct March 2025","provider":"Allen AI","country":"US","openness":"open","release_date":"2025-03-13","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":11,"benchmark_coverage_score":9.13,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.37,"source_primary":"crfm.stanford.edu","source_primary_share":0.4545,"source_primary_group_share":0.4545,"capability_breadth":2.28,"capability_primary":"instruction_following","capability_primary_share":0.5455,"capability_primary_group_count":6,"capability_non_primary_group_count":5},{"model":"Marin 8B Instruct","provider":"Other","country":"XX","openness":"open","release_date":"2025-05-14","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":11,"benchmark_coverage_score":9.13,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.37,"source_primary":"crfm.stanford.edu","source_primary_share":0.4545,"source_primary_group_share":0.4545,"capability_breadth":2.28,"capability_primary":"instruction_following","capability_primary_share":0.5455,"capability_primary_group_count":6,"capability_non_primary_group_count":5},{"model":"OLMo 2 13B Instruct November 2024","provider":"Allen AI","country":"US","openness":"open","release_date":"2024-11-26","param_count":13.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":11,"benchmark_coverage_score":9.13,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.37,"source_primary":"crfm.stanford.edu","source_primary_share":0.4545,"source_primary_group_share":0.4545,"capability_breadth":2.28,"capability_primary":"instruction_following","capability_primary_share":0.5455,"capability_primary_group_count":6,"capability_non_primary_group_count":5},{"model":"OLMo 2 7B Instruct November 2024","provider":"Allen AI","country":"US","openness":"open","release_date":"2024-11-26","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":11,"benchmark_coverage_score":9.13,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.37,"source_primary":"crfm.stanford.edu","source_primary_share":0.4545,"source_primary_group_share":0.4545,"capability_breadth":2.28,"capability_primary":"instruction_following","capability_primary_share":0.5455,"capability_primary_group_count":6,"capability_non_primary_group_count":5},{"model":"OLMoE 1B-7B Instruct January 2025","provider":"Allen AI","country":"US","openness":"open","release_date":"2025-01-27","param_count":1.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":11,"benchmark_coverage_score":9.13,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.37,"source_primary":"crfm.stanford.edu","source_primary_share":0.4545,"source_primary_group_share":0.4545,"capability_breadth":2.28,"capability_primary":"instruction_following","capability_primary_share":0.5455,"capability_primary_group_count":6,"capability_non_primary_group_count":5},{"model":"Gemini 3 Flash (Preview) (Medium)","provider":"Google","country":"US","openness":"api","release_date":"2025-12-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":4.0,"source_primary":"contextarena.ai","source_primary_share":0.3,"source_primary_group_share":0.375,"capability_breadth":5.0,"capability_primary":"longctx","capability_primary_share":0.3,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"recurrentgemma-9B-it","provider":"Google","country":"US","openness":"api","release_date":"2024-06-07","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.5455,"source_primary_group_share":0.75,"capability_breadth":3.46,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4545,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"MN-12B-Mag-Mell-R1","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-16","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.17,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.6,"capability_breadth":4.17,"capability_primary":"knowledge_general","capability_primary_share":0.4,"capability_primary_group_count":4,"capability_non_primary_group_count":6},{"model":"internlm2-chat-1.8B","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2024-01-30","param_count":1.8,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.5455,"source_primary_group_share":0.75,"capability_breadth":3.46,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4545,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"SmolLM-1.7B","provider":"HuggingFace","country":"FR","openness":"open","release_date":"2024-07-14","param_count":1.7,"active_param_count":null,"training_compute_flop":1.03e+22,"num_benchmarks":11,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.8571,"capability_breadth":3.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"OLMo-1B-hf","provider":"Allen AI","country":"US","openness":"open","release_date":"2024-04-12","param_count":1.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.8571,"capability_breadth":3.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"SmolLM-360M","provider":"HuggingFace","country":"FR","openness":"open","release_date":"2024-07-14","param_count":0.36,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.8571,"capability_breadth":3.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"SmolLM-135M","provider":"HuggingFace","country":"FR","openness":"open","release_date":"2024-07-14","param_count":0.135,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.8571,"capability_breadth":3.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"SmolLM-360M-Instruct","provider":"HuggingFace","country":"FR","openness":"open","release_date":"2024-07-15","param_count":0.36,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.5455,"source_primary_group_share":0.75,"capability_breadth":4.48,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3636,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"Claude Opus 4.1 (Non-reasoning)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-08-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":6.25,"source_primary":"huggingface.co/spaces/dontplantoend/ugi-leaderboard","source_primary_share":0.3,"source_primary_group_share":0.3,"capability_breadth":4.55,"capability_primary":"knowledge_general","capability_primary_share":0.3,"capability_primary_group_count":3,"capability_non_primary_group_count":7},{"model":"Claude Opus 4.1 (High)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-08-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":9,"benchmark_coverage_score":7.45,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.25,"source_primary":"hal.cs.princeton.edu","source_primary_share":0.9091,"source_primary_group_share":0.8889,"capability_breadth":1.86,"capability_primary":"agentic","capability_primary_share":0.6364,"capability_primary_group_count":5,"capability_non_primary_group_count":4},{"model":"GPT-5 Mini (Low)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-08-07","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":11,"benchmark_coverage_score":11.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":4.84,"source_primary":"huggingface.co/spaces/aiml-tuda/slr-leaderboard","source_primary_share":0.3636,"source_primary_group_share":0.3636,"capability_breadth":3.46,"capability_primary":"reasoning_logical","capability_primary_share":0.4545,"capability_primary_group_count":5,"capability_non_primary_group_count":6},{"model":"Kimi K3 (Max)","provider":"Moonshot","country":"CN","openness":"api","release_date":"2026-07-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":11,"benchmark_coverage_score":11.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.28,"source_primary":"epoch.ai","source_primary_share":0.6364,"source_primary_group_share":0.6364,"capability_breadth":4.84,"capability_primary":"math","capability_primary_share":0.3636,"capability_primary_group_count":4,"capability_non_primary_group_count":7},{"model":"ERNIE 4.5","provider":"Baidu","country":"CN","openness":"api","release_date":"2025-03-16","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":2,"source_effective_breadth":1.88,"source_primary":"fab-ai.org","source_primary_share":0.5455,"source_primary_group_share":0.625,"capability_breadth":2.57,"capability_primary":"vision_language","capability_primary_share":0.5455,"capability_primary_group_count":5,"capability_non_primary_group_count":3},{"model":"Qwen3-VL-235B-A22B-Thinking-FP8","provider":"Alibaba","country":"CN","openness":"open","release_date":"2025-09-22","param_count":235.0,"active_param_count":22.0,"training_compute_flop":null,"num_benchmarks":11,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.69,"source_primary":"github.com/geralt-targaryen/mc-evaluation","source_primary_share":0.5455,"source_primary_group_share":0.7143,"capability_breadth":2.2,"capability_primary":"math","capability_primary_share":0.6364,"capability_primary_group_count":3,"capability_non_primary_group_count":4},{"model":"DeepSeek V2 Chat","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2024-05-06","param_count":236.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":10,"benchmark_coverage_score":9.52,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":3.33,"source_primary":"huggingface.co/spaces/eduagarcia/open_pt_llm_leaderboard","source_primary_share":0.5,"source_primary_group_share":0.5,"capability_breadth":2.38,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6,"capability_primary_group_count":6,"capability_non_primary_group_count":4},{"model":"Nxcode-CQ-7B-orpo","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-24","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.5,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.6,"capability_breadth":5.0,"capability_primary":"code","capability_primary_share":0.3,"capability_primary_group_count":3,"capability_non_primary_group_count":7},{"model":"deepseek-llm-67B-base","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2023-11-29","param_count":67.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.59,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8,"source_primary_group_share":0.7778,"capability_breadth":2.5,"capability_primary":"knowledge_general","capability_primary_share":0.6,"capability_primary_group_count":5,"capability_non_primary_group_count":4},{"model":"Baichuan-2-13B-Base","provider":"Baichuan","country":"CN","openness":"open","release_date":"2023-09-06","param_count":13.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.25,"source_primary":"epoch.ai","source_primary_share":0.9,"source_primary_group_share":0.8889,"capability_breadth":5.56,"capability_primary":"knowledge_general","capability_primary_share":0.3,"capability_primary_group_count":3,"capability_non_primary_group_count":6},{"model":"O3 Mini (2025-01-31) (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-01-31","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":4.76,"source_primary":"github.com/lmarena/arena-hard-auto","source_primary_share":0.3,"source_primary_group_share":0.3333,"capability_breadth":5.0,"capability_primary":"creativity","capability_primary_share":0.3,"capability_primary_group_count":2,"capability_non_primary_group_count":7},{"model":"PaLM-2 (Bison)","provider":"Google","country":"US","openness":"api","release_date":"2023-05-10","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google PaLM family retired or superseded by Gemini models.","num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":4,"source_group_family_count":3,"source_effective_breadth":2.31,"source_primary":"crfm.stanford.edu","source_primary_share":0.5,"source_primary_group_share":0.5556,"capability_breadth":2.5,"capability_primary":"knowledge_general","capability_primary_share":0.6,"capability_primary_group_count":5,"capability_non_primary_group_count":4},{"model":"Claude Instant 1.1","provider":"Anthropic","country":"US","openness":"api","release_date":"2023-11-22","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 1/2 family retired or superseded by newer Claude models.","num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.98,"source_primary":"epoch.ai","source_primary_share":0.5556,"source_primary_group_share":0.5556,"capability_breadth":3.0,"capability_primary":"structured_precision","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"code-davinci-002","provider":"OpenAI","country":"US","openness":"api","release_date":"2022-03-15","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.22,"source_primary":"crfm.stanford.edu","source_primary_share":0.9,"source_primary_group_share":0.9,"capability_breadth":2.94,"capability_primary":"math","capability_primary_share":0.4,"capability_primary_group_count":4,"capability_non_primary_group_count":6},{"model":"OLMo-7B-Instruct","provider":"Allen AI","country":"US","openness":"open","release_date":"2024-02-23","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.5,"source_primary":"huggingface.co","source_primary_share":0.6,"source_primary_group_share":0.6,"capability_breadth":4.17,"capability_primary":"reasoning_logical","capability_primary_share":0.4,"capability_primary_group_count":4,"capability_non_primary_group_count":6},{"model":"gemma-2-9B-it-DPO","provider":"Google","country":"US","openness":"api","release_date":"2024-07-16","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.5,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.6,"capability_breadth":5.0,"capability_primary":"instruction_following","capability_primary_share":0.3,"capability_primary_group_count":3,"capability_non_primary_group_count":7},{"model":"Starling-LM-7B-beta-ExPO","provider":"UC Berkeley","country":"US","openness":"open","release_date":"2024-04-26","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.59,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8,"source_primary_group_share":0.7778,"capability_breadth":3.12,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"MAmmoTH2-8x7B-Plus","provider":"CMU","country":"US","openness":"open","release_date":"2024-05-06","param_count":56.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":7.69,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.59,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8,"source_primary_group_share":0.7778,"capability_breadth":1.92,"capability_primary":"knowledge_general","capability_primary_share":0.7,"capability_primary_group_count":6,"capability_non_primary_group_count":3},{"model":"Chocolatine-2-14B-Instruct-v2.0","provider":"Other","country":"XX","openness":"open","release_date":"2025-01-23","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.75,"capability_breadth":4.17,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"Virtuoso-Lite","provider":"Arcee AI","country":"US","openness":"open","release_date":"2025-01-28","param_count":10.3,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.75,"capability_breadth":4.17,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"calme-3.3-instruct-3B","provider":"Other","country":"XX","openness":"open","release_date":"2024-11-07","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.75,"capability_breadth":4.17,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"calme-3.3-baguette-3B","provider":"Other","country":"XX","openness":"open","release_date":"2024-11-07","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.75,"capability_breadth":4.17,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"calme-3.2-baguette-3B","provider":"Other","country":"XX","openness":"open","release_date":"2024-11-07","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.75,"capability_breadth":4.17,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"calme-3.1-baguette-3B","provider":"Other","country":"XX","openness":"open","release_date":"2024-11-07","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.75,"capability_breadth":4.17,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"EnnoAi-7B-French-Instruct-202502","provider":"Other","country":"XX","openness":"open","release_date":"2025-02-11","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.75,"capability_breadth":4.17,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"calme-3.1-instruct-3B","provider":"Other","country":"XX","openness":"open","release_date":"2024-11-07","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.75,"capability_breadth":4.17,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"Napoleon_24B_V0.0","provider":"Other","country":"XX","openness":"open","release_date":"2025-02-16","param_count":23.6,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.75,"capability_breadth":4.17,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"Pensez-Llama3.1-8B","provider":"Other","country":"XX","openness":"open","release_date":"2025-02-19","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.75,"capability_breadth":4.17,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"Napoleon_24B_V0.2","provider":"Other","country":"XX","openness":"open","release_date":"2025-02-17","param_count":23.6,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.75,"capability_breadth":4.17,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"calme-3.1-llamaloi-3B","provider":"Other","country":"XX","openness":"open","release_date":"2024-11-07","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.75,"capability_breadth":4.17,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"ReasoningCore-3B-RE1-V2","provider":"Other","country":"XX","openness":"open","release_date":"2025-02-18","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.75,"capability_breadth":4.17,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"ReasoningCore-3B-T1.1","provider":"Other","country":"XX","openness":"open","release_date":"2025-02-06","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.75,"capability_breadth":4.17,"capability_primary":"knowledge_multilingual","capability_primary_share":0.3,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"EXAONE-3.5-2.4B-Instruct","provider":"LG AI","country":"KR","openness":"open","release_date":"2024-12-01","param_count":2.4,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.5,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.6,"capability_breadth":4.17,"capability_primary":"knowledge_general","capability_primary_share":0.4,"capability_primary_group_count":4,"capability_non_primary_group_count":6},{"model":"ko-gemma-2-9B-it","provider":"Other","country":"XX","openness":"open","release_date":"2024-07-11","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":7.69,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"lk.instruct.kr","source_primary_share":0.9,"source_primary_group_share":0.875,"capability_breadth":1.92,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7,"capability_primary_group_count":5,"capability_non_primary_group_count":3},{"model":"palm-2-codechat-bison-32k","provider":"Google","country":"US","openness":"api","release_date":"2023-05-10","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google PaLM family retired or superseded by Gemini models.","num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"lk.instruct.kr","source_primary_share":0.9,"source_primary_group_share":0.875,"capability_breadth":2.38,"capability_primary":"knowledge_multilingual","capability_primary_share":0.6,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"EEVE-Korean-Instruct-10.8B-v1.0","provider":"Yanolja","country":"KR","openness":"open","release_date":"2024-02-22","param_count":10.8,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":7.69,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"lk.instruct.kr","source_primary_share":0.9,"source_primary_group_share":0.875,"capability_breadth":1.92,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7,"capability_primary_group_count":5,"capability_non_primary_group_count":3},{"model":"KULLM3","provider":"Korea University","country":"KR","openness":"open","release_date":"2024-04-02","param_count":10.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":7.69,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"lk.instruct.kr","source_primary_share":0.9,"source_primary_group_share":0.875,"capability_breadth":1.92,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7,"capability_primary_group_count":5,"capability_non_primary_group_count":3},{"model":"Bookworm-10.7B-v0.4-DPO","provider":"Other","country":"XX","openness":"open","release_date":"2024-01-18","param_count":10.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":7.69,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"lk.instruct.kr","source_primary_share":0.9,"source_primary_group_share":0.875,"capability_breadth":1.92,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7,"capability_primary_group_count":5,"capability_non_primary_group_count":3},{"model":"Synatra-7B-v0.3-dpo","provider":"Other","country":"XX","openness":"open","release_date":"2023-11-08","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":7.69,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"lk.instruct.kr","source_primary_share":0.9,"source_primary_group_share":0.875,"capability_breadth":1.92,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7,"capability_primary_group_count":5,"capability_non_primary_group_count":3},{"model":"AlphaMist7B-slr-v4-slow2","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-10","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":7.69,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"lk.instruct.kr","source_primary_share":0.9,"source_primary_group_share":0.875,"capability_breadth":1.92,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7,"capability_primary_group_count":5,"capability_non_primary_group_count":3},{"model":"T3Q-ko-solar-dpo-v3.0","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-20","param_count":10.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":7.69,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"lk.instruct.kr","source_primary_share":0.9,"source_primary_group_share":0.875,"capability_breadth":1.92,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7,"capability_primary_group_count":5,"capability_non_primary_group_count":3},{"model":"EEVE-Korean-10.8B-instruct-v4-sft","provider":"Yanolja","country":"KR","openness":"open","release_date":"2024-07-02","param_count":10.8,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":7.69,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"lk.instruct.kr","source_primary_share":0.9,"source_primary_group_share":0.875,"capability_breadth":1.92,"capability_primary":"knowledge_multilingual","capability_primary_share":0.7,"capability_primary_group_count":5,"capability_non_primary_group_count":3},{"model":"openchat_3.5","provider":"Other","country":"XX","openness":"open","release_date":"2023-10-30","param_count":7.2,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.5,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.6,"capability_breadth":5.56,"capability_primary":"knowledge_general","capability_primary_share":0.3,"capability_primary_group_count":3,"capability_non_primary_group_count":7},{"model":"Gemini 3.1 Flash Lite (Preview) (High)","provider":"Google","country":"US","openness":"api","release_date":"2026-05-07","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Google lists gemini-3.1-flash-lite-preview with a May 25, 2026 shutdown date and recommends gemini-3.1-flash-lite.","num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.59,"source_primary":"benchmarklist.com","source_primary_share":0.8,"source_primary_group_share":0.7778,"capability_breadth":3.33,"capability_primary":"code","capability_primary_share":0.4,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"doubao-seed-1.6-thinking-250615","provider":"ByteDance","country":"CN","openness":"api","release_date":"2025-06-25","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.98,"source_primary":"rank.opencompass.org.cn","source_primary_share":0.7,"source_primary_group_share":0.6667,"capability_breadth":3.57,"capability_primary":"knowledge_general","capability_primary_share":0.4,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Gemini 2.5 Flash (Preview 04-17) (Thinking)","provider":"Google","country":"US","openness":"api","release_date":"2025-04-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 2.5 preview snapshot retired or superseded by newer Gemini models.","num_benchmarks":10,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.38,"source_primary":"vals.ai","source_primary_share":0.6,"source_primary_group_share":0.6,"capability_breadth":5.0,"capability_primary":"math","capability_primary_share":0.3,"capability_primary_group_count":3,"capability_non_primary_group_count":7},{"model":"solar-pro-preview-instruct","provider":"Upstage","country":"KR","openness":"open","release_date":"2024-09-09","param_count":22.1,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.8571,"capability_breadth":3.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"zetasepic-abliteratedV2-Qwen2.5-32B-Inst-BaseMerge-TIES","provider":"Other","country":"XX","openness":"open","release_date":"2024-12-07","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Qwen2.5-14B-Instruct-abliterated-v2","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-10-09","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Dans-PersonalityEngine-V1.2.0-24B","provider":"Other","country":"XX","openness":"open","release_date":"2025-02-17","param_count":24.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Dobby-Mini-Leashed-Llama-3.1-8B","provider":"Other","country":"XX","openness":"open","release_date":"2025-01-22","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Dobby-Mini-Unhinged-Llama-3.1-8B","provider":"Other","country":"XX","openness":"open","release_date":"2025-01-22","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Wingless_Imp_8B","provider":"Other","country":"XX","openness":"open","release_date":"2025-01-24","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"L3-70B-Euryale-v2.1","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-11","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"BaeZel-8B-LINEAR","provider":"Other","country":"XX","openness":"open","release_date":"2024-11-08","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Tiger-Gemma-9B-v1","provider":"Other","country":"XX","openness":"open","release_date":"2024-07-12","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Llama-3.1-Nemotron-lorablated-70B","provider":"NVIDIA","country":"US","openness":"open","release_date":"2024-10-17","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Aspire-8B-model_stock","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-16","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Dans-PersonalityEngine-V1.1.0-12B","provider":"Other","country":"XX","openness":"open","release_date":"2024-12-19","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Tiger-Gemma-9B-v2","provider":"Other","country":"XX","openness":"open","release_date":"2024-07-16","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"L3-8B-Lunaris-v1","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-26","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"L3-8B-Stheno-v3.2","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-05","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Tiger-Gemma-9B-v3","provider":"Other","country":"XX","openness":"open","release_date":"2024-10-04","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"N3N_gemma-2-9B-it_20241029_1532","provider":"Other","country":"XX","openness":"open","release_date":"2024-10-29","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"MS-Meadowlark-22B","provider":"Other","country":"XX","openness":"open","release_date":"2024-10-18","param_count":22.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"70B-L3.3-Cirrus-x1","provider":"Other","country":"XX","openness":"open","release_date":"2025-01-06","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"gemma2-9B-cpt-sahabatai-v1-instruct","provider":"Other","country":"XX","openness":"open","release_date":"2024-11-06","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.75,"capability_breadth":3.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"Dans-SakuraKaze-V1.0.0-12B","provider":"Other","country":"XX","openness":"open","release_date":"2025-02-13","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Llama 3.2 3B Mix Skill","provider":"Meta","country":"US","openness":"open","release_date":"2024-10-24","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"L3-DARKEST-PLANET-16.5B","provider":"Other","country":"XX","openness":"open","release_date":"2024-10-11","param_count":16.5,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Explore_Llama-3.2-1B-Inst_v1.1","provider":"Other","country":"XX","openness":"open","release_date":"2024-10-09","param_count":1.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"MSH-v1-Bielik-v2.3-Instruct-MedIT-merge","provider":"Other","country":"XX","openness":"open","release_date":"2024-10-29","param_count":11.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.8571,"capability_breadth":3.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"StarDust-12B-v2","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-01","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"magnum-v4-22B","provider":"Other","country":"XX","openness":"open","release_date":"2024-10-20","param_count":22.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"ASTAROTH-3.2-1B","provider":"Other","country":"XX","openness":"open","release_date":"2025-03-09","param_count":1.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Arcee-Blitz","provider":"Arcee AI","country":"US","openness":"open","release_date":"2025-02-07","param_count":24.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"La_Mejor_Mezcla-3.2-1B","provider":"Other","country":"XX","openness":"open","release_date":"2025-03-03","param_count":1.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Dolphin3.0-Llama3.2-1B","provider":"Cognitive Computations","country":"US","openness":"open","release_date":"2024-12-29","param_count":1.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"MN-LooseCannon-12B-v1","provider":"Other","country":"XX","openness":"open","release_date":"2024-08-09","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"gemma-2-2B-it-abliterated","provider":"Google","country":"US","openness":"api","release_date":"2024-06-27","param_count":2.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Llama 3 8B Cpt Sahabatai V1 Instruct","provider":"Meta","country":"US","openness":"open","release_date":null,"param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.75,"capability_breadth":3.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"mini-magnum-12B-v1.1","provider":"Other","country":"XX","openness":"open","release_date":"2024-07-24","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Fimbulvetr-11B-v2","provider":"Other","country":"XX","openness":"open","release_date":"2024-02-06","param_count":11.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Lumimaid-v0.2-8B","provider":"Other","country":"XX","openness":"open","release_date":"2024-07-24","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Captain_BMO-12B","provider":"Other","country":"XX","openness":"open","release_date":"2024-11-01","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Herodotos-14B","provider":"Other","country":"XX","openness":"open","release_date":"2025-01-13","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"ReWiz-Llama-3.2-3B","provider":"Other","country":"XX","openness":"open","release_date":"2024-10-18","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Violet_Twilight-v0.2","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-12","param_count":12.2,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Llama-3SOME-8B-v2","provider":"Meta","country":"US","openness":"open","release_date":"2024-06-21","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"RYS-Phi-3-medium-4k-instruct","provider":"Other","country":"XX","openness":"open","release_date":"2024-08-06","param_count":17.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.8571,"capability_breadth":3.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"2B-ad","provider":"Other","country":"XX","openness":"open","release_date":"2024-08-26","param_count":2.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Captain-Eris_Violet-V0.420-12B","provider":"Other","country":"XX","openness":"open","release_date":"2024-12-16","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"patricide-12B-Unslop-Mell","provider":"Other","country":"XX","openness":"open","release_date":"2024-12-01","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"BLAST_PROCESSING-3.2-1B","provider":"Other","country":"XX","openness":"open","release_date":"2025-03-06","param_count":1.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"SOLAR-10.7B-Instruct-v1.0-uncensored","provider":"Upstage","country":"KR","openness":"open","release_date":"2023-12-14","param_count":10.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"magnum-v2-12B","provider":"Other","country":"XX","openness":"open","release_date":"2024-08-03","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"MSH-Lite-7B-v1-Bielik-v2.3-Instruct-Llama-Prune","provider":"Other","country":"XX","openness":"open","release_date":"2024-11-12","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.8571,"capability_breadth":3.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Dusk_Rainbow","provider":"Other","country":"XX","openness":"open","release_date":"2024-08-16","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Hermes-3-Llama-3.1-70B-lorablated","provider":"Nous Research","country":"US","openness":"open","release_date":"2024-08-16","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Evil-Alpaca-3B-L3.2","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-28","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Impish_Mind_8B","provider":"Other","country":"XX","openness":"open","release_date":"2024-12-05","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Gemma-2-Ataraxy-9B","provider":"Other","country":"XX","openness":"open","release_date":"2024-08-14","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Magnum-Picaro-0.7-v2-12B","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-11","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Gemmasutra-Mini-2B-v1","provider":"Other","country":"XX","openness":"open","release_date":"2024-08-03","param_count":2.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"h2o-danube3-4B-base","provider":"H2O.ai","country":"US","openness":"open","release_date":"2024-07-04","param_count":4.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.8571,"capability_breadth":3.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Minitron-4B-Base","provider":"NVIDIA","country":"US","openness":"open","release_date":"2024-07-19","param_count":4.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.8571,"capability_breadth":3.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"internlm2-1.8B","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2024-01-30","param_count":1.8,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.8571,"capability_breadth":3.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"mt5-xl","provider":"Google","country":"US","openness":"api","release_date":"2020-10-22","param_count":3.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.8571,"capability_breadth":3.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"MT-Merge6-gemma-2-9B","provider":"Other","country":"XX","openness":"open","release_date":"2025-02-13","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"MT-Gen6fix-gemma-2-9B","provider":"Other","country":"XX","openness":"open","release_date":"2025-02-02","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Lumimaid-v0.2-12B","provider":"Other","country":"XX","openness":"open","release_date":"2024-07-25","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.8,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.52,"capability_primary":"knowledge_general","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"TinyLlama-1.1B-Chat-v1.0","provider":"Singapore Univ","country":"SG","openness":"open","release_date":"2023-12-30","param_count":1.1,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6,"source_primary_group_share":0.8571,"capability_breadth":3.85,"capability_primary":"knowledge_multilingual","capability_primary_share":0.4,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"MiniCPM-SALA","provider":"OpenBMB","country":"CN","openness":"open","release_date":"2026-02-11","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.98,"source_primary":"rank.opencompass.org.cn","source_primary_share":0.7,"source_primary_group_share":0.6667,"capability_breadth":3.12,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":5,"capability_non_primary_group_count":4},{"model":"NVIDIA-Nemotron-3-Super-120B-A12B-BF16","provider":"NVIDIA","country":"US","openness":"open","release_date":"2026-03-11","param_count":120.0,"active_param_count":12.0,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":2.0,"source_primary":"huggingface.co/spaces/hebrew-llm-leaderboard/leaderboard","source_primary_share":0.5,"source_primary_group_share":0.5,"capability_breadth":2.78,"capability_primary":"knowledge_multilingual","capability_primary_share":0.5,"capability_primary_group_count":5,"capability_non_primary_group_count":5},{"model":"uitars-1.5-7B","provider":"Other","country":"XX","openness":"open","release_date":null,"param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":10,"benchmark_coverage_score":10.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.12,"source_primary":"huggingface.co/spaces/astarforbae/scratchworld-leaderboard","source_primary_share":0.5,"source_primary_group_share":0.5,"capability_breadth":3.85,"capability_primary":"agentic","capability_primary_share":0.4,"capability_primary_group_count":4,"capability_non_primary_group_count":6},{"model":"GLM-5.2 (High)","provider":"Zhipu","country":"CN","openness":"open","release_date":"2026-06-13","param_count":744.0,"active_param_count":40.0,"training_compute_flop":null,"num_benchmarks":10,"benchmark_group_count":10,"benchmark_coverage_score":7.41,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":7.14,"source_primary":"aikido.dev","source_primary_share":0.2,"source_primary_group_share":0.2,"capability_breadth":1.85,"capability_primary":"code","capability_primary_share":0.7,"capability_primary_group_count":7,"capability_non_primary_group_count":3},{"model":"dolphin-2.6-mixtral-8x7B","provider":"Cognitive Computations","country":"US","openness":"open","release_date":"2023-12-21","param_count":56.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.79,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"deepseek-coder-1.3B-base","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2023-11-02","param_count":1.3,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":2.79,"source_primary":"epoch.ai","source_primary_share":0.5556,"source_primary_group_share":0.5556,"capability_breadth":3.52,"capability_primary":"code","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"CodeLlama-34B-Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2023-08-24","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":5.44,"source_primary":"scale.com","source_primary_share":0.2857,"source_primary_group_share":0.2857,"capability_breadth":1.81,"capability_primary":"code","capability_primary_share":0.7143,"capability_primary_group_count":5,"capability_non_primary_group_count":2},{"model":"Baichuan-2-7B-Base","provider":"Baichuan","country":"CN","openness":"open","release_date":"2023-09-06","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"epoch.ai","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":4.76,"capability_primary":"knowledge_general","capability_primary_share":0.3333,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"llama-v3p3-70B-instruct","provider":"Meta","country":"US","openness":"open","release_date":"2024-12-06","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.25,"source_primary":"huggingface.co/spaces/gsma/open-telco-llm-leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.8889,"capability_breadth":2.61,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":5,"capability_non_primary_group_count":4},{"model":"O1 Preview (2024-09-12)","provider":"OpenAI","country":"US","openness":"api","release_date":"2024-09-12","param_count":200.0,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy OpenAI preview model superseded by later o-series models.","num_benchmarks":9,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":7.36,"source_primary":"epoch.ai","source_primary_share":0.2222,"source_primary_group_share":0.2222,"capability_breadth":4.76,"capability_primary":"code","capability_primary_share":0.3333,"capability_primary_group_count":3,"capability_non_primary_group_count":6},{"model":"Palmyra X V3 (72B)","provider":"Writer","country":"US","openness":"api","release_date":"2024-01-01","param_count":72.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":2,"source_effective_breadth":1.88,"source_primary":"crfm.stanford.edu","source_primary_share":0.5556,"source_primary_group_share":0.625,"capability_breadth":2.08,"capability_primary":"knowledge_general","capability_primary_share":0.6667,"capability_primary_group_count":5,"capability_non_primary_group_count":3},{"model":"PaLM-2 (Unicorn)","provider":"Google","country":"US","openness":"api","release_date":"2023-05-10","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google PaLM family retired or superseded by Gemini models.","num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":2,"source_effective_breadth":1.88,"source_primary":"crfm.stanford.edu","source_primary_share":0.5556,"source_primary_group_share":0.625,"capability_breadth":2.08,"capability_primary":"knowledge_general","capability_primary_share":0.6667,"capability_primary_group_count":5,"capability_non_primary_group_count":3},{"model":"Palmyra X V2 (33B)","provider":"Writer","country":"US","openness":"api","release_date":"2023-12-01","param_count":33.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":2,"source_effective_breadth":1.88,"source_primary":"crfm.stanford.edu","source_primary_share":0.5556,"source_primary_group_share":0.625,"capability_breadth":2.08,"capability_primary":"knowledge_general","capability_primary_share":0.6667,"capability_primary_group_count":5,"capability_non_primary_group_count":3},{"model":"Nous-Hermes-2-Yi-34B","provider":"Nous Research","country":"US","openness":"open","release_date":"2023-12-23","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.79,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"Nanbeige2-16B-Chat","provider":"Nanbeige","country":"CN","openness":"open","release_date":"2024-05-24","param_count":16.0,"active_param_count":null,"training_compute_flop":4.05e+23,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.79,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"internlm2-chat-20B-ExPO","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2024-05-02","param_count":20.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.79,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"tulu-2-dpo-70B-ExPO","provider":"Allen AI","country":"US","openness":"open","release_date":"2024-04-26","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.79,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"internlm2-chat-7B-ExPO","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2024-05-02","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.79,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"Starling-LM-7B-alpha-ExPO","provider":"UC Berkeley","country":"US","openness":"open","release_date":"2024-04-26","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.79,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"XuanYuan-70B","provider":"Du Xiaoman","country":"CN","openness":"open","release_date":"2023-09-22","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.08,"capability_primary":"knowledge_general","capability_primary_share":0.6667,"capability_primary_group_count":5,"capability_non_primary_group_count":3},{"model":"Faro-Yi-9B","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-27","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.61,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"multi_verse_model","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-07","param_count":7.2,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.61,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"GeM2-Llamion-14B-Chat","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-13","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.61,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"GeM2-Llamion-14B-Base","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-13","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.61,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"GeM2-Llamion-14B-LongChat","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-29","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.61,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"llama-3-nectar-dpo-8B","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-14","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.61,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"Configurable-Llama-3-8B-v0.3","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-20","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.61,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"NeuralLLaMa-3-8B-ORPO-v0.4","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-16","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.61,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"Tess-2.0-Llama-3-8B","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-05","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.61,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"openchat-3.5-0106-laser","provider":"Other","country":"XX","openness":"open","release_date":"2024-01-27","param_count":7.2,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.61,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"juud-Mistral-7B-dpo","provider":"Other","country":"XX","openness":"open","release_date":"2024-02-07","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.61,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"Configurable-Mistral-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-11","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.61,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"janus-orpo-7B","provider":"KAIST","country":"KR","openness":"open","release_date":"2024-04-18","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.61,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"Synatra-7B-v0.3-RP","provider":"Other","country":"XX","openness":"open","release_date":"2023-10-29","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.61,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"dolphin-2.9.1-mixtral-1x22B","provider":"Cognitive Computations","country":"US","openness":"open","release_date":"2024-05-22","param_count":22.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.61,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"openbuddy-mistral-22B-v21.1-32k","provider":"Other","country":"XX","openness":"open","release_date":"2024-05-17","param_count":22.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"huggingface.co/spaces/baai/open_cn_llm_leaderboard","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":2.61,"capability_primary":"knowledge_general","capability_primary_share":0.5556,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"GPT-5.4 (Thinking)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-03-06","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.29,"source_primary":"deploymentsafety.openai.com","source_primary_share":0.625,"source_primary_group_share":0.625,"capability_breadth":4.0,"capability_primary":"knowledge_general","capability_primary_share":0.375,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"Llama-3.1-Tulu-3-70B-DPO","provider":"Meta","country":"US","openness":"open","release_date":"2024-11-20","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.98,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.86,"capability_primary":"math","capability_primary_share":0.3333,"capability_primary_group_count":3,"capability_non_primary_group_count":6},{"model":"Llama 3.2 11B Vision Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2024-09-25","param_count":11.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.08,"source_primary":"benchmarklist.com","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.86,"capability_primary":"vision_language","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Mistral: Mistral Small 3.1 24B","provider":"Mistral","country":"FR","openness":"open","release_date":null,"param_count":24.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.53,"source_primary":"benchmarklist.com","source_primary_share":0.7778,"source_primary_group_share":0.7778,"capability_breadth":5.4,"capability_primary":"math","capability_primary_share":0.3333,"capability_primary_group_count":3,"capability_non_primary_group_count":6},{"model":"Llama 3.1 8B IT","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-23","param_count":8.0,"active_param_count":null,"training_compute_flop":1.224e+24,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"balrogai.com","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":2.13,"capability_primary":"reasoning_planning","capability_primary_share":0.625,"capability_primary_group_count":5,"capability_non_primary_group_count":3},{"model":"Gemini 2.0 Flash Lite (Preview)","provider":"Google","country":"US","openness":"api","release_date":"2025-02-25","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 2.0 family deprecated in favor of newer Gemini models.","num_benchmarks":9,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.08,"source_primary":"artificialanalysis.ai","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.24,"capability_primary":"code","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Claude Opus 4.5 (20251101) (Thinking 32K)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-11-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.46,"source_primary":"epoch.ai","source_primary_share":0.4444,"source_primary_group_share":0.5,"capability_breadth":3.52,"capability_primary":"code","capability_primary_share":0.4444,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"DeepSeek V4 Pro (High)","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2026-04-23","param_count":1600.0,"active_param_count":49.0,"training_compute_flop":9.702e+24,"num_benchmarks":9,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":7.36,"source_primary":"yhenon.github.io","source_primary_share":0.2222,"source_primary_group_share":0.2222,"capability_breadth":3.52,"capability_primary":"code","capability_primary_share":0.3333,"capability_primary_group_count":3,"capability_non_primary_group_count":6},{"model":"Claude Opus 4 (20250514) (Thinking)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-05-14","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 4.0 family deprecated in favor of newer Claude 4.x models.","num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":4.0,"source_primary":"huggingface.co/spaces/dontplantoend/ugi-leaderboard","source_primary_share":0.375,"source_primary_group_share":0.375,"capability_breadth":3.56,"capability_primary":"knowledge_general","capability_primary_share":0.375,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"Claude Sonnet 5 (Max)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-06-30","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.08,"source_primary":"epoch.ai","source_primary_share":0.6667,"source_primary_group_share":0.6667,"capability_breadth":3.0,"capability_primary":"math","capability_primary_share":0.4444,"capability_primary_group_count":4,"capability_non_primary_group_count":5},{"model":"Phi-lthy4","provider":"Microsoft","country":"US","openness":"open","release_date":"2025-02-12","param_count":11.9,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":2.91,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"Phi-Line_14B","provider":"Microsoft","country":"US","openness":"open","release_date":"2025-02-17","param_count":14.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":2.91,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"Redemption_Wind_24B","provider":"Other","country":"XX","openness":"open","release_date":"2025-02-06","param_count":23.6,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":2.91,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"2B_or_not_2B","provider":"Other","country":"XX","openness":"open","release_date":"2024-08-11","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":2.91,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"Claude Sonnet 4.5 (Thinking 16K)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-09-29","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":6.23,"source_primary":"arcprize.org","source_primary_share":0.2222,"source_primary_group_share":0.2222,"capability_breadth":4.26,"capability_primary":"reasoning_logical","capability_primary_share":0.3333,"capability_primary_group_count":3,"capability_non_primary_group_count":6},{"model":"Gemini 3.1 Pro (Preview) (Low)","provider":"Google","country":"US","openness":"api","release_date":"2026-02-19","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":5.44,"source_primary":"contextarena.ai","source_primary_share":0.3333,"source_primary_group_share":0.2857,"capability_breadth":4.26,"capability_primary":"longctx","capability_primary_share":0.3333,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"mistral-small-latest","provider":"Mistral","country":"FR","openness":"open","release_date":"2026-03-16","param_count":119.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.46,"source_primary":"huggingface.co","source_primary_share":0.5,"source_primary_group_share":0.5,"capability_breadth":2.91,"capability_primary":"code","capability_primary_share":0.5,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"mistral-small-creative","provider":"Mistral","country":"FR","openness":"open","release_date":null,"param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":4.0,"source_primary":"huggingface.co","source_primary_share":0.375,"source_primary_group_share":0.375,"capability_breadth":2.13,"capability_primary":"creativity","capability_primary_share":0.625,"capability_primary_group_count":5,"capability_non_primary_group_count":3},{"model":"Kimi K2 Turbo (Thinking)","provider":"Moonshot","country":"CN","openness":"open","release_date":"2025-11-06","param_count":1000.0,"active_param_count":32.0,"training_compute_flop":2.976e+24,"num_benchmarks":9,"benchmark_group_count":9,"benchmark_coverage_score":9.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":3.0,"source_primary":"epoch.ai","source_primary_share":0.4444,"source_primary_group_share":0.4444,"capability_breadth":4.76,"capability_primary":"code","capability_primary_share":0.3333,"capability_primary_group_count":3,"capability_non_primary_group_count":6},{"model":"Nemotron-4 15B","provider":"NVIDIA","country":"US","openness":"open","release_date":"2024-03-01","param_count":15.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":9,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"epoch.ai","source_primary_share":0.8889,"source_primary_group_share":0.875,"capability_breadth":4.26,"capability_primary":"reasoning_logical","capability_primary_share":0.3333,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"Qwen-1.8B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2023-11-30","param_count":1.8,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"epoch.ai","source_primary_share":0.875,"source_primary_group_share":0.875,"capability_breadth":4.57,"capability_primary":"knowledge_general","capability_primary_share":0.375,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"OpenHermes-2-Mistral-7B","provider":"Other","country":"XX","openness":"open","release_date":"2023-10-12","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":5.33,"capability_primary":"reasoning_logical","capability_primary_share":0.25,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"Yi-6B-200K","provider":"01.AI","country":"CN","openness":"open","release_date":"2023-11-06","param_count":6.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":4.57,"capability_primary":"longctx","capability_primary_share":0.25,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"Storm-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-25","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":4.57,"capability_primary":"instruction_following","capability_primary_share":0.25,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"Llama-3-Instruct-8B-SimPO-ExPO","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-26","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":3.56,"capability_primary":"instruction_following","capability_primary_share":0.375,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"Infinity-Instruct-7M-Gen-mistral-7B","provider":"BAAI","country":"CN","openness":"open","release_date":"2024-07-25","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":4.57,"capability_primary":"instruction_following","capability_primary_share":0.25,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"Infinity-Instruct-3M-0625-Mistral-7B","provider":"BAAI","country":"CN","openness":"open","release_date":"2024-07-09","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":4.57,"capability_primary":"instruction_following","capability_primary_share":0.25,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"Infinity-Instruct-3M-0613-Mistral-7B","provider":"BAAI","country":"CN","openness":"open","release_date":"2024-06-21","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":4.57,"capability_primary":"instruction_following","capability_primary_share":0.25,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"Infinity-Instruct-3M-0625-Qwen2-7B","provider":"BAAI","country":"CN","openness":"open","release_date":"2024-07-09","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":4.57,"capability_primary":"instruction_following","capability_primary_share":0.25,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"opt-13B","provider":"Meta","country":"US","openness":"open","release_date":"2022-05-03","param_count":13.0,"active_param_count":null,"training_compute_flop":3.53e+23,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"epoch.ai","source_primary_share":0.875,"source_primary_group_share":0.875,"capability_breadth":2.91,"capability_primary":"knowledge_general","capability_primary_share":0.5,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"OpenThinker-7B","provider":"Other","country":"XX","openness":"open","release_date":"2025-01-28","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":3.56,"capability_primary":"knowledge_general","capability_primary_share":0.375,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"Llama 3 8B Magpie Align V0.1","provider":"Meta","country":"US","openness":"open","release_date":"2024-06-29","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":4.57,"capability_primary":"reasoning_logical","capability_primary_share":0.25,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"Llama-3-Instruct-8B-SimPO-v0.2","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-06","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":4.57,"capability_primary":"reasoning_logical","capability_primary_share":0.25,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"Gemma-2-9B-It-SPPO-Iter3","provider":"Google","country":"US","openness":"api","release_date":"2024-06-29","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":4.57,"capability_primary":"knowledge_multilingual","capability_primary_share":0.25,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"NuminaMath-7B-CoT","provider":"Other","country":"XX","openness":"open","release_date":"2024-07-15","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":4.57,"capability_primary":"math","capability_primary_share":0.25,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"gte-Qwen2-7B-instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-06-15","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Kimi-VL-A3B","provider":"Moonshot","country":"CN","openness":"open","release_date":"2025-04-10","param_count":null,"active_param_count":2.8,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":5,"source_group_family_count":5,"source_effective_breadth":3.2,"source_primary":"huggingface.co/spaces/aiml-tuda/slr-leaderboard","source_primary_share":0.5,"source_primary_group_share":0.5,"capability_breadth":2.0,"capability_primary":"vision_language","capability_primary_share":0.5,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"GPT-5.2 Codex (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2026-01-14","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.58,"source_primary":"mercor.com","source_primary_share":0.5714,"source_primary_group_share":0.5714,"capability_breadth":3.27,"capability_primary":"knowledge_general","capability_primary_share":0.4286,"capability_primary_group_count":3,"capability_non_primary_group_count":4},{"model":"O3 Pro (High)","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-06-10","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":5.33,"source_primary":"arcprize.org","source_primary_share":0.25,"source_primary_group_share":0.25,"capability_breadth":2.67,"capability_primary":"vision_language","capability_primary_share":0.5,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"Claude Opus 4.6 (Thinking, High)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-02-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":8,"source_group_family_count":8,"source_effective_breadth":8.0,"source_primary":"github.com/lechmazur/buyout_game","source_primary_share":0.125,"source_primary_group_share":0.125,"capability_breadth":5.33,"capability_primary":"reasoning_planning","capability_primary_share":0.25,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"Qwen 3 Max (2025-09-23)","provider":"Alibaba","country":"CN","openness":"api","release_date":"2025-09-23","param_count":700.0,"active_param_count":null,"training_compute_flop":1.512e+25,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.29,"source_primary":"epoch.ai","source_primary_share":0.625,"source_primary_group_share":0.625,"capability_breadth":4.0,"capability_primary":"knowledge_general","capability_primary_share":0.375,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"Qwen2.5-Math-1.5B-Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-16","param_count":1.5,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.6,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":3.56,"capability_primary":"math","capability_primary_share":0.375,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"granite-3.2-8B-instruct","provider":"IBM","country":"US","openness":"open","release_date":"2025-02-26","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":5.33,"capability_primary":"reasoning_logical","capability_primary_share":0.25,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"zephyr-orpo-141B-A35B-v0.1","provider":"HuggingFace","country":"FR","openness":"open","release_date":"2024-04-10","param_count":141.0,"active_param_count":35.0,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.68,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.75,"source_primary_group_share":0.75,"capability_breadth":4.0,"capability_primary":"knowledge_general","capability_primary_share":0.375,"capability_primary_group_count":3,"capability_non_primary_group_count":5},{"model":"Claude Sonnet 4 (Thinking 16K)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-05-22","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Anthropic Claude 4.0 family deprecated in favor of newer Claude 4.x models.","num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":6.4,"source_primary":"arcprize.org","source_primary_share":0.25,"source_primary_group_share":0.25,"capability_breadth":3.2,"capability_primary":"reasoning_logical","capability_primary_share":0.5,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"GLM-4.7 (Thinking)","provider":"Zhipu","country":"CN","openness":"open","release_date":"2025-12-22","param_count":355.0,"active_param_count":32.0,"training_compute_flop":4.42e+24,"num_benchmarks":8,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":7.0,"source_primary":"raw.githubusercontent.com/llm2014/llm_benchmark","source_primary_share":0.25,"source_primary_group_share":0.1429,"capability_breadth":3.2,"capability_primary":"knowledge_general","capability_primary_share":0.375,"capability_primary_group_count":3,"capability_non_primary_group_count":4},{"model":"Gemini 3 Flash (Preview) (Minimal)","provider":"Google","country":"US","openness":"api","release_date":"2025-12-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":3.27,"source_primary":"huggingface.co/spaces/dontplantoend/ugi-leaderboard","source_primary_share":0.4286,"source_primary_group_share":0.4286,"capability_breadth":5.44,"capability_primary":"knowledge_general","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"mistral-large-latest","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-07-24","param_count":123.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":3.27,"source_primary":"github.com/ukgovernmentbeis/inspect_evals","source_primary_share":0.4286,"source_primary_group_share":0.4286,"capability_breadth":2.33,"capability_primary":"code","capability_primary_share":0.5714,"capability_primary_group_count":4,"capability_non_primary_group_count":3},{"model":"DeepSeek V4 Pro (Reasoning)","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2026-04-23","param_count":1600.0,"active_param_count":49.0,"training_compute_flop":9.702e+24,"num_benchmarks":8,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":3.27,"source_primary":"huggingface.co/spaces/dontplantoend/ugi-leaderboard","source_primary_share":0.4286,"source_primary_group_share":0.4286,"capability_breadth":4.45,"capability_primary":"knowledge_multilingual","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"GLM-5.2 (Reasoning)","provider":"Zhipu","country":"CN","openness":"open","release_date":"2026-06-13","param_count":744.0,"active_param_count":40.0,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":3.27,"source_primary":"huggingface.co/spaces/dontplantoend/ugi-leaderboard","source_primary_share":0.4286,"source_primary_group_share":0.4286,"capability_breadth":4.45,"capability_primary":"knowledge_multilingual","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"DeepSeek V4 Flash (Reasoning)","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2026-04-23","param_count":284.0,"active_param_count":13.0,"training_compute_flop":2.496e+24,"num_benchmarks":8,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":3.27,"source_primary":"huggingface.co/spaces/dontplantoend/ugi-leaderboard","source_primary_share":0.4286,"source_primary_group_share":0.4286,"capability_breadth":4.45,"capability_primary":"knowledge_multilingual","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Claude Sonnet 5 (xHigh)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-06-30","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.29,"source_primary":"epoch.ai","source_primary_share":0.625,"source_primary_group_share":0.625,"capability_breadth":6.4,"capability_primary":"code","capability_primary_share":0.25,"capability_primary_group_count":2,"capability_non_primary_group_count":6},{"model":"LFM 40B","provider":"Liquid AI","country":"US","openness":"open","release_date":"2024-10-23","param_count":40.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":8,"benchmark_group_count":8,"benchmark_coverage_score":8.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.28,"source_primary":"artificialanalysis.ai","source_primary_share":0.875,"source_primary_group_share":0.875,"capability_breadth":2.46,"capability_primary":"code","capability_primary_share":0.5,"capability_primary_group_count":4,"capability_non_primary_group_count":4},{"model":"finma-7B-full","provider":"Other","country":"XX","openness":"open","release_date":"2023-06-13","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.81,"source_primary":"huggingface.co/spaces/thefinai/open-finllm-leaderboard","source_primary_share":0.7143,"source_primary_group_share":0.7143,"capability_breadth":1.81,"capability_primary":"knowledge_general","capability_primary_share":0.7143,"capability_primary_group_count":5,"capability_non_primary_group_count":2},{"model":"Claude Opus 4.6 (Thinking 32K)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-02-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"epoch.ai","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":3.77,"capability_primary":"math","capability_primary_share":0.4286,"capability_primary_group_count":3,"capability_non_primary_group_count":4},{"model":"speechless-codellama-34B-v2.0","provider":"Other","country":"XX","openness":"open","release_date":"2023-10-04","param_count":34.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"speechless-coder-ds-6.7B","provider":"Other","country":"XX","openness":"open","release_date":"2023-12-30","param_count":6.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Tess-v2.5.2-Qwen2-72B","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-13","param_count":72.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"deepseek-moe-16B-chat","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2024-01-09","param_count":16.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"deepseek-moe-16B-base","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2024-01-08","param_count":16.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Qwen2-Math-72B-Instruct","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-08-08","param_count":72.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":3.77,"capability_primary":"math","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Cydonia-22B-v1.2","provider":"Other","country":"XX","openness":"open","release_date":"2024-10-07","param_count":22.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":3.77,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Nous-Hermes-2-Mixtral-8x7B-SFT","provider":"Nous Research","country":"US","openness":"open","release_date":"2023-12-26","param_count":56.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Yarn-Mistral-7B-128k","provider":"Other","country":"XX","openness":"open","release_date":"2023-10-31","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"gemma-2-9B-it-WPO-HB","provider":"Google","country":"US","openness":"api","release_date":"2024-08-08","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":3.77,"capability_primary":"instruction_following","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"FuseChat-Llama-3.1-8B-Instruct","provider":"Other","country":"XX","openness":"open","release_date":"2024-11-20","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":3.77,"capability_primary":"instruction_following","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"FuseChat-Qwen-2.5-7B-Instruct","provider":"Other","country":"XX","openness":"open","release_date":"2024-11-12","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":3.77,"capability_primary":"instruction_following","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"FuseChat-Llama-3.2-3B-Instruct","provider":"Other","country":"XX","openness":"open","release_date":"2024-12-06","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":3.77,"capability_primary":"instruction_following","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Infinity-Instruct-7M-Gen-Llama3.1-70B","provider":"BAAI","country":"CN","openness":"open","release_date":"2024-07-25","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":3.77,"capability_primary":"instruction_following","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Infinity-Instruct-3M-0625-Llama3-70B","provider":"BAAI","country":"CN","openness":"open","release_date":"2024-07-09","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":3.77,"capability_primary":"instruction_following","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Infinity-Instruct-3M-0613-Llama3-70B","provider":"BAAI","country":"CN","openness":"open","release_date":"2024-06-27","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":3.77,"capability_primary":"instruction_following","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Nanda-10B-Chat","provider":"Meta","country":"US","openness":"open","release_date":"2024-10-22","param_count":10.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"AryaBhatta-GemmaOrca-Merged","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-01","param_count":8.5,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"AryaBhatta-GemmaUltra-Merged","provider":"Other","country":"XX","openness":"open","release_date":"2024-04-12","param_count":8.5,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Airavata","provider":"AI4Bharat","country":"IN","openness":"open","release_date":"2024-01-13","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"SELM-Llama-3-8B-Instruct-iter-3","provider":"Other","country":"XX","openness":"open","release_date":"2024-08-06","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":3.77,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3.1-OpenScholar-8B","provider":"Meta","country":"US","openness":"open","release_date":"2024-11-15","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":3.27,"capability_primary":"knowledge_general","capability_primary_share":0.4286,"capability_primary_group_count":3,"capability_non_primary_group_count":4},{"model":"neural-chat-7B-v3-3","provider":"Other","country":"XX","openness":"open","release_date":"2023-12-09","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Gemma-2-Ataraxy-Gemmasutra-9B-slerp","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-11","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Gemma-2-9B-It-SPPO-Iter1","provider":"Google","country":"US","openness":"api","release_date":"2024-06-29","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Gemma-2-9B-It-SPPO-Iter2","provider":"Google","country":"US","openness":"api","release_date":"2024-06-29","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"qwen2.5-reinstruct-alternate-lumen-14B","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-23","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Qwen2.5-14B-Gutenberg-Instruct-Slerpeno","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-20","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"gemma2-gutenberg-27B","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-09","param_count":27.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"recoilme-gemma-2-9B-v0.3","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-18","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"gemma-2-27B-it-SimPO-37K","provider":"Google","country":"US","openness":"api","release_date":"2024-08-13","param_count":27.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"romulus-mistral-nemo-12B-simpo","provider":"Other","country":"XX","openness":"open","release_date":"2024-07-24","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"gemma-2-27B-it-SimPO-37K-100steps","provider":"Google","country":"US","openness":"api","release_date":"2024-08-13","param_count":27.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"mistral-nemo-wissenschaft-12B","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-08-12","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"magnum-v4-9B","provider":"Other","country":"XX","openness":"open","release_date":"2024-10-20","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Odin-9B","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-27","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"magnum-v3-27B-kto","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-06","param_count":27.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Gemma2-9B-IT-Simpo-Infinity-Preference","provider":"Other","country":"XX","openness":"open","release_date":"2024-08-28","param_count":9.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"dolphin-2.9.3-mistral-nemo-12B","provider":"Cognitive Computations","country":"US","openness":"open","release_date":"2024-07-23","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"T3Q-Qwen2.5-7B-it-KOR-Safe","provider":"Other","country":"XX","openness":"open","release_date":"2024-11-11","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Base-8B-SFT-RDPO","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-17","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Chocolatine-14B-Instruct-4k-DPO","provider":"Other","country":"XX","openness":"open","release_date":"2024-08-01","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3.1-SuperNova-8B-Lite_TIES_with_Base","provider":"Meta","country":"US","openness":"open","release_date":"2024-10-02","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Chocolatine-14B-Instruct-DPO-v1.2","provider":"Other","country":"XX","openness":"open","release_date":"2024-08-12","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Ph3della5-14B","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-06","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Instruct-8B-CPO-v0.2","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-06","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Instruct-8B-ORPO-v0.2","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-06","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Instruct-8B-CPO","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-06","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Base-8B-SFT-SLiC-HF","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-06","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama 3.1 8B Ultra Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2024-08-26","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Meta-Llama-3.1-8B-Instruct-HalfAbliterated-TIES","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-19","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Instruct-8B-DPO","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-17","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Tess-3-Mistral-Nemo-12B","provider":"Other","country":"XX","openness":"open","release_date":"2024-08-13","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Instruct-8B-SPPO-Iter1","provider":"Meta","country":"US","openness":"open","release_date":"2024-06-25","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Trinity-2-Codestral-22B","provider":"Arcee AI","country":"US","openness":"open","release_date":"2024-08-07","param_count":22.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama 3.1 8B MultiReflection Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2024-09-15","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3.1-SauerkrautLM-8B-Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-25","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Instruct-8B-ORPO","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-17","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Base-8B-SFT-ORPO","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-17","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Base-8B-SFT-KTO","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-17","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Fireball-Alpaca-Llama3.1.07-8B-Philos-Math-KTO-beta","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-12","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"calme-2.2-qwen2-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-27","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Defne-llama3.1-8B","provider":"Other","country":"XX","openness":"open","release_date":"2024-07-29","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Instruct-8B-SLiC-HF","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-06","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Instruct-8B-RRHF","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-06","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"L3.1-Niitorm-8B-DPO-t0.0001","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-19","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Instruct-8B-SLiC-HF-v0.2","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-06","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Base-8B-SFT-DPO","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-17","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"openbuddy-llama3.1-8B-v22.3-131k","provider":"Other","country":"XX","openness":"open","release_date":"2024-08-16","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"calme-2.1-qwen2-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-27","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"calme-2.3-qwen2-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-27","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Meta-Llama-3.1-8B-Instruct-abliterated","provider":"Other","country":"XX","openness":"open","release_date":"2024-07-24","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"HyperLlama-3.1-8B","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-04","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"ConfigurableSOLAR-10.7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-03-10","param_count":10.7,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Einstein-v7-Qwen2-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-24","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Trinity-2-Codestral-22B-v0.2","provider":"Arcee AI","country":"US","openness":"open","release_date":"2024-08-13","param_count":22.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Instruct-8B-RDPO-v0.2","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-06","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3.2-Rabbit-Ko-3B-Instruct","provider":"Meta","country":"US","openness":"open","release_date":"2024-09-30","param_count":3.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3.1-8B-Instruct-abliterated_via_adapter","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-25","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Instruct-8B-DPO-v0.2","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-06","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3.1-8B-Lexi-Uncensored","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-26","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"openbuddy-qwen2.5llamaify-14B-v23.1-200k","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-23","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Base-8B-SFT-IPO","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-17","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Mistral-7B-Instruct-SLiC-HF","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-07-06","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"naps-llama-3.1-instruct-v0.5.0","provider":"Other","country":"XX","openness":"open","release_date":null,"param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama 3.1 8B Cobalt","provider":"Meta","country":"US","openness":"open","release_date":"2024-08-16","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Mistral-7B-Instruct-ORPO","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-05-17","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Mistral-7B-Base-SFT-CPO","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-07-06","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Mistral-7B-Instruct-CPO","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-07-06","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"llama-3-Korean-8B","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-10","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"AnFeng_v3.1-Avocet","provider":"Other","country":"XX","openness":"open","release_date":"2024-08-03","param_count":34.4,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Qwen2-7B-Multilingual-RP","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-06-24","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Mistral-7B-Base-SFT-KTO","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-05-17","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"dolphin-2.9.2-qwen2-7B","provider":"Cognitive Computations","country":"US","openness":"open","release_date":"2024-05-24","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"naps-llama-3.1-8B-instruct-v0.4","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-12","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Mistral-7B-Instruct-RRHF","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-07-06","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"calme-2.5-qwen2-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-27","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Instruct-8B-RDPO","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-17","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Mistral7B-PairRM-SPPO-Iter1","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-05-04","param_count":7.2,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Base-8B-SFT-RRHF","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-06","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Instruct-8B-KTO","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-17","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"RYS-Llama-3.1-8B-Instruct","provider":"Other","country":"XX","openness":"open","release_date":"2024-08-08","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"naps-llama-3.1-8B-instruct-v0.3","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-02","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Mistral-7B-Base-SFT-SimPO","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-05-17","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Mistral-7B-Instruct-IPO","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-05-17","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Mistral-7B-Instruct-SimPO","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-05-24","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"calme-2.4-qwen2-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-27","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Mistral-7B-Base-SFT-SLiC-HF","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-07-06","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Mistral-7B-Base-SFT-RDPO","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-05-17","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"naps-llama-3.1_instruct-v0.6.0","provider":"Other","country":"XX","openness":"open","release_date":null,"param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"calme-2.6-qwen2-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-27","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Samantha-Qwen-2-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-15","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Mistral-7B-Instruct-KTO","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-05-17","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Mistral-7B-Base-SFT-IPO","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-05-17","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Mistral7B-PairRM-SPPO","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-05-04","param_count":7.2,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama 3.1 8B Fireplace2","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-23","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Mistral-7B-Base-SFT-RRHF","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-07-06","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Base-8B-SFT","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-17","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"llama-3-Korean-8B-r-v-0.1","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-18","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Mistral-7B-Instruct-RDPO","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-05-17","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Phi-3-medium-4k-instruct-abliterated-v3","provider":"Microsoft","country":"US","openness":"open","release_date":"2024-05-22","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama 3.1 8B ShiningValiant2","provider":"Meta","country":"US","openness":"open","release_date":"2024-08-06","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"dolphin-2.9.3-mistral-7B-32k","provider":"Cognitive Computations","country":"US","openness":"open","release_date":"2024-06-25","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Base-8B-SFT-SimPO","provider":"Meta","country":"US","openness":"open","release_date":"2024-05-24","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Mistral-7B-Instruct-DPO","provider":"Mistral","country":"FR","openness":"open","release_date":null,"param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Mistral-7B-Base-SFT-DPO","provider":"Mistral","country":"FR","openness":"open","release_date":"2024-05-17","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Instruct-8B-KTO-v0.2","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-06","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"openbuddy-qwen2.5llamaify-7B-v23.1-200k","provider":"Other","country":"XX","openness":"open","release_date":"2024-10-04","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"calme-2.7-qwen2-7B","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-27","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"MagpieLM-8B-SFT-v0.1","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-15","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Gemma-Ko-Merge","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-04","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama 3 8B Magpie Align SFT V0.1","provider":"Meta","country":"US","openness":"open","release_date":"2024-06-06","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama 3.1 8B Magpie Align V0.1","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-24","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"MagpieLM-8B-Chat-v0.1","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-15","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama 3.1 8B Magpie Align SFT V0.1","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-23","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-3-Base-8B-SFT-CPO","provider":"Meta","country":"US","openness":"open","release_date":"2024-07-06","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Qwen2-7B-Instruct-abliterated","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-06-14","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Tess-v2.5-Phi-3-medium-128k-14B","provider":"Other","country":"XX","openness":"open","release_date":"2024-06-05","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"NuminaMath-7B-TIR","provider":"Other","country":"XX","openness":"open","release_date":"2024-07-04","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"dolphin-2.9.4-llama3.1-8B","provider":"Cognitive Computations","country":"US","openness":"open","release_date":"2024-08-04","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Echo","provider":"Other","country":"XX","openness":"open","release_date":null,"param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Fino1-8B","provider":"Other","country":"XX","openness":"open","release_date":"2025-02-12","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.69,"source_primary":"huggingface.co/spaces/thefinai/open-finllm-reasoning-leaderboard","source_primary_share":0.7143,"source_primary_group_share":0.7143,"capability_breadth":2.33,"capability_primary":"math","capability_primary_share":0.5714,"capability_primary_group_count":4,"capability_non_primary_group_count":3},{"model":"Fino1-14B","provider":"Other","country":"XX","openness":"open","release_date":"2025-03-28","param_count":14.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.69,"source_primary":"huggingface.co/spaces/thefinai/open-finllm-reasoning-leaderboard","source_primary_share":0.7143,"source_primary_group_share":0.7143,"capability_breadth":2.33,"capability_primary":"math","capability_primary_share":0.5714,"capability_primary_group_count":4,"capability_non_primary_group_count":3},{"model":"MiMo-V2-Pro (Hunter Alpha)","provider":"Xiaomi","country":"CN","openness":"open","release_date":"2026-03-11","param_count":1000.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":5.44,"source_primary":"eqbench.com","source_primary_share":0.2857,"source_primary_group_share":0.2857,"capability_breadth":3.27,"capability_primary":"creativity","capability_primary_share":0.4286,"capability_primary_group_count":3,"capability_non_primary_group_count":4},{"model":"DeepSeek V3.2 (High)","provider":"DeepSeek","country":"CN","openness":"open","release_date":"2025-12-01","param_count":671.0,"active_param_count":37.0,"training_compute_flop":4.2e+24,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":3.27,"source_primary":"benchmarklist.com","source_primary_share":0.4286,"source_primary_group_share":0.4286,"capability_breadth":7.0,"capability_primary":"reasoning_planning","capability_primary_share":0.1429,"capability_primary_group_count":1,"capability_non_primary_group_count":6},{"model":"Llama3-Med42-70B","provider":"Meta","country":"US","openness":"open","release_date":"2024-06-27","param_count":70.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":3.27,"capability_primary":"knowledge_general","capability_primary_share":0.4286,"capability_primary_group_count":3,"capability_non_primary_group_count":4},{"model":"Claude Sonnet 4.6 (Thinking, High)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-02-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":7.0,"source_primary":"github.com/lechmazur/buyout_game","source_primary_share":0.1429,"source_primary_group_share":0.1429,"capability_breadth":4.45,"capability_primary":"reasoning_planning","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Claude Opus 4.6 (Medium)","provider":"Anthropic","country":"US","openness":"api","release_date":"2026-02-05","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":7,"source_group_family_count":7,"source_effective_breadth":7.0,"source_primary":"chessbenchllm.onrender.com","source_primary_share":0.1429,"source_primary_group_share":0.1429,"capability_breadth":2.58,"capability_primary":"reasoning_planning","capability_primary_share":0.4286,"capability_primary_group_count":3,"capability_non_primary_group_count":4},{"model":"Claude Opus 4.5 (20251101) (Unknown)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-11-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.58,"source_primary":"epoch.ai","source_primary_share":0.5714,"source_primary_group_share":0.5714,"capability_breadth":3.77,"capability_primary":"agentic","capability_primary_share":0.4286,"capability_primary_group_count":3,"capability_non_primary_group_count":4},{"model":"Claude Sonnet 4.5 (Unknown)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-09-29","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":2.58,"source_primary":"epoch.ai","source_primary_share":0.5714,"source_primary_group_share":0.5714,"capability_breadth":3.27,"capability_primary":"agentic","capability_primary_share":0.4286,"capability_primary_group_count":3,"capability_non_primary_group_count":4},{"model":"EXAONE 3.5 32B Instruct","provider":"LG AI","country":"KR","openness":"open","release_date":"2024-12-09","param_count":32.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"EXAONE 3.5 7.8B Instruct","provider":"LG AI","country":"KR","openness":"open","release_date":"2024-12-01","param_count":7.8,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"internlm2.5-20B-chat","provider":"Shanghai AI Lab","country":"CN","openness":"open","release_date":"2024-07-30","param_count":20.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":3.27,"capability_primary":"knowledge_general","capability_primary_share":0.4286,"capability_primary_group_count":3,"capability_non_primary_group_count":4},{"model":"granite-3.2-2B-instruct","provider":"IBM","country":"US","openness":"open","release_date":"2025-02-17","param_count":2.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Vikhr-Nemo-12B-Instruct-R-21-09-24","provider":"Other","country":"XX","openness":"open","release_date":"2024-09-21","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"SphinX","provider":"Other","country":"XX","openness":"open","release_date":"2023-11-02","param_count":13.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"K2-Chat","provider":"LLM360","country":"UAE","openness":"open","release_date":"2024-05-22","param_count":65.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":3.27,"capability_primary":"knowledge_general","capability_primary_share":0.4286,"capability_primary_group_count":3,"capability_non_primary_group_count":4},{"model":"DeepHermes-3-Mistral-24B-Preview","provider":"Nous Research","country":"US","openness":"open","release_date":"2025-03-02","param_count":24.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Cotype-Nano","provider":"Other","country":"XX","openness":"open","release_date":null,"param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"OLMo-7B-Instruct-hf","provider":"Allen AI","country":"US","openness":"open","release_date":"2024-06-04","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama-PLLuM-8B-base","provider":"Meta","country":"US","openness":"open","release_date":"2025-02-07","param_count":8.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"PLLuM-12B-base","provider":"PLLuM","country":"PL","openness":"open","release_date":"2025-02-07","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"PLLuM-12B-nc-base","provider":"PLLuM","country":"PL","openness":"open","release_date":"2025-02-07","param_count":12.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"codegemma-1.1-2B","provider":"Google","country":"US","openness":"api","release_date":"2024-04-30","param_count":2.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Yarn-Llama-2-7B-64k","provider":"Other","country":"XX","openness":"open","release_date":"2023-08-30","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Yarn-Llama-2-7B-128k","provider":"Other","country":"XX","openness":"open","release_date":"2023-08-31","param_count":7.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"huggingface.co/spaces/open-llm-leaderboard/open_llm_leaderboard","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Sarvam 105B","provider":"Sarvam AI","country":"IN","openness":"open","release_date":"2026-02-18","param_count":105.0,"active_param_count":9.0,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.33,"source_primary":"llm-stats.com","source_primary_share":0.5714,"source_primary_group_share":0.5714,"capability_breadth":2.58,"capability_primary":"math","capability_primary_share":0.5714,"capability_primary_group_count":4,"capability_non_primary_group_count":3},{"model":"Sarvam 30B","provider":"Sarvam AI","country":"IN","openness":"api","release_date":"2026-02-18","param_count":30.0,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":2.33,"source_primary":"llm-stats.com","source_primary_share":0.5714,"source_primary_group_share":0.5714,"capability_breadth":2.58,"capability_primary":"math","capability_primary_share":0.5714,"capability_primary_group_count":4,"capability_non_primary_group_count":3},{"model":"GPT-5.2 ChatGPT","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-12-11","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":5.44,"source_primary":"dubesor.de","source_primary_share":0.2857,"source_primary_group_share":0.2857,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"GPT-5.1 ChatGPT","provider":"OpenAI","country":"US","openness":"api","release_date":"2025-11-12","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":6,"source_group_family_count":6,"source_effective_breadth":5.44,"source_primary":"dubesor.de","source_primary_share":0.2857,"source_primary_group_share":0.2857,"capability_breadth":4.45,"capability_primary":"reasoning_logical","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5},{"model":"Llama 3.3 70B IT","provider":"Meta","country":"US","openness":"open","release_date":"2024-12-06","param_count":70.0,"active_param_count":null,"training_compute_flop":6.8649768e+24,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"balrogai.com","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":1.81,"capability_primary":"reasoning_planning","capability_primary_share":0.7143,"capability_primary_group_count":5,"capability_non_primary_group_count":2},{"model":"Qwen 2.5 7B (IT)","provider":"Alibaba","country":"CN","openness":"open","release_date":"2024-09-19","param_count":7.0,"active_param_count":null,"training_compute_flop":8.219e+23,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":2,"source_group_family_count":2,"source_effective_breadth":1.32,"source_primary":"balrogai.com","source_primary_share":0.8571,"source_primary_group_share":0.8571,"capability_breadth":1.81,"capability_primary":"reasoning_planning","capability_primary_share":0.7143,"capability_primary_group_count":5,"capability_non_primary_group_count":2},{"model":"Gemini 2.5 Flash (Preview)","provider":"Google","country":"US","openness":"api","release_date":"2025-06-17","param_count":null,"active_param_count":null,"training_compute_flop":null,"lifecycle":"deprecated","lifecycle_reason":"Legacy Google Gemini 2.5 preview snapshot retired or superseded by newer Gemini models.","num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":4,"source_group_family_count":4,"source_effective_breadth":3.77,"source_primary":"arcprize.org","source_primary_share":0.2857,"source_primary_group_share":0.2857,"capability_breadth":3.27,"capability_primary":"vision_language","capability_primary_share":0.4286,"capability_primary_group_count":3,"capability_non_primary_group_count":4},{"model":"Claude Haiku 4.5 (20251001) (Unknown)","provider":"Anthropic","country":"US","openness":"api","release_date":"2025-10-01","param_count":null,"active_param_count":null,"training_compute_flop":null,"num_benchmarks":7,"benchmark_group_count":7,"benchmark_coverage_score":7.0,"source_family_count":3,"source_group_family_count":3,"source_effective_breadth":1.81,"source_primary":"epoch.ai","source_primary_share":0.7143,"source_primary_group_share":0.7143,"capability_breadth":4.45,"capability_primary":"code","capability_primary_share":0.2857,"capability_primary_group_count":2,"capability_non_primary_group_count":5}]}