EvalPlus (HumanEval+ & MBPP+): leaderboard
EvalPlus leaderboard — HumanEval+ and MBPP+ code generation with vastly expanded test suites over the originals; pass@1 correctness under rigorous testing.
Metric: Pass@1 avg (%). Source: evalplus.github.io. Saturation forecast: Estimated already saturated. 125 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4 Turbo | 86.6 | #404 |
| 2 | O1 Preview (2024-09-12) | 84.6 | #197 |
| 3 | O1 Mini (2024-09-12) | 83.9 | #403 |
| 4 | Qwen 2.5 Coder 32B Instruct | 82.1 | #528 |
| 5 | GPT-4o (2024-08-06) | 79.7 | #336 |
| 6 | GPT-4 | 79.3 | #438 |
| 7 | Claude 3.5 Sonnet (20240620) | 78 | #299 |
| 8 | GPT-4o Mini (2024-07-18) | 77.8 | #558 |
| 9 | GPT-4 Turbo (Preview) | 77.5 | #397 |
| 10 | Gemini 1.5 Pro (002) | 76.9 | #349 |
| 11 | Claude 3 Opus | 75.3 | #571 |
| 12 | OpenCoder-8B-Instruct | 74.4 | #1338 |
| 13 | CodeQwen1.5-7B Chat | 73.8 | #1144 |
| 14 | Grok Beta | 73 | #487 |
| 15 | Gemini 1.5 Flash (002) | 71.5 | #583 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=evalplus-humaneval-plus-mbpp-plus · How It Works · Data refreshed daily, snapshot 2026-10-20.