EvoEval Difficult — leaderboard

Metric: Pass@1 (%). Source: evo-eval.github.io. 17 models tracked.

Top models

#ModelScore
1GPT-452
2GPT-4 Turbo50
3Claude 3 Haiku40
4deepseek-coder-6.7B-instruct40
5Gemini 1.0 Pro37
6Claude 229
7Mixtral 8x7B Instruct21
8deepseek-coder-1.3B-instruct20
9Phi-218
10starcoder2-15B16
11gemma-7B12
12starcoder12
13starcoder2-7B12
14Mistral 7B8
15starcoder2-3B8

Interactive version: aibenchmarks.dev/benchmark?slug=evoeval-difficult · How the rankings work · Data refreshed daily, snapshot 2026-07-08.