LLM Hallucination Detection - RAG — leaderboard
Metric: Hallucination rate (%). Source: huggingface.co. 15 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Mistral Nemo Instruct (2407) | 10.63 |
| 2 | Qwen 2.5 VL 7B Instruct | 9.35 |
| 3 | Magistral Small | 8.62 |
| 4 | Qwen 3 235B A22B 2507 | 7.53 |
| 5 | Qwen 3 235B A22B FP8 | 6.16 |
| 6 | GPT-4o | 6.05 |
| 7 | Mistral Small 3 | 4.74 |
| 8 | DeepSeek V3 (0324) | 4.66 |
| 9 | Llama 4 Scout Instruct | 4.23 |
| 10 | Gemma 3 27B (IT) | 3.71 |
| 11 | Llama 4 Maverick Instruct FP8 | 3.34 |
| 12 | Kimi K2 | 2.9 |
| 13 | DeepSeek R1 0528 | 2.26 |
| 14 | Claude Sonnet 4 | 2.21 |
| 15 | Gemini 2.5 Pro | 1.57 |
Interactive version: aibenchmarks.dev/benchmark?slug=llm-hallucination-detection-rag · How the rankings work · Data refreshed daily, snapshot 2026-07-08.