EvalPlus (HumanEval+ / MBPP+): leaderboard
Enhanced versions of HumanEval and MBPP with 80x and 35x more test cases respectively. Measures Pass@k for single-function code generation.
Metric: Pass@1 avg (%). Source: evalplus.github.io. 125 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-4 Turbo | 86.6 |
| 2 | O1 Preview (2024-09-12) | 84.6 |
| 3 | O1 Mini (2024-09-12) | 83.9 |
| 4 | Qwen 2.5 Coder 32B Instruct | 82.1 |
| 5 | GPT-4o (2024-08-06) | 79.7 |
| 6 | GPT-4 | 79.3 |
| 7 | Claude 3.5 Sonnet (20240620) | 78 |
| 8 | GPT-4o Mini (2024-07-18) | 77.8 |
| 9 | GPT-4 Turbo (Preview) | 77.5 |
| 10 | Gemini 1.5 Pro (002) | 76.9 |
| 11 | Claude 3 Opus | 75.3 |
| 12 | OpenCoder-8B-Instruct | 74.4 |
| 13 | CodeQwen1.5-7B Chat | 73.8 |
| 14 | Grok Beta | 73 |
| 15 | Gemini 1.5 Flash (002) | 71.5 |
Interactive version: theaggregate.ai/benchmark?slug=evalplus-humaneval-plus-mbpp-plus-1e45jvt · How It Works · Data refreshed daily, snapshot 2026-10-09.