OpenAI's HumanEval benchmark — 164 hand-crafted Python programming problems testing function-level code generation from docstrings.
Foundational coding benchmark. Well-studied with extensive baselines. Tests core Python programming ability.
Largely saturated (top models >95%). Problems are in training data. Only tests Python. Short, isolated functions don't reflect real software engineering.
Higher is better. Pass@1 percentage (0–100%). Frontier models score >95%.
Largely saturated (top models >95%).
| # | Model | Score |
|---|---|---|
| 1 | 98.5 | |
| 2 | 97.6 | |
| 3 | 97.6 | |
| 4 | 96.7 | |
| 5 | 96.6 | |
| 6 | 96.2 | |
| 7 | 96.1 | |
| 8 | 96.1 | |
| 9 | 96.1 | |
| 10 | 95.9 | |
| 11 | 95.9 | |
| 12 | 95.8 | |
| 13 | 95.7 | |
| 14 | 95.3 | |
| 15 | 95.3 | |
| 16 | 95.3 | |
| 17 | 95.1 | |
| 18 | 95.1 | |
| 19 | 95.0 | |
| 20 | 95.0 | |
| 21 | 95.0 | |
| 22 | 95.0 | |
| 23 | 95.0 | |
| 24 | 95.0 | |
| 25 | 95.0 | |
| 26 | 95.0 | |
| 27 | 95.0 | |
| 28 | 95.0 | |
| 29 | 95.0 | |
| 30 | 95.0 | |
| 31 | 94.9 | |
| 32 | 94.9 | |
| 33 | 94.9 | |
| 34 | 94.8 | |
| 35 | 94.7 | |
| 36 | 94.7 | |
| 37 | 94.7 | |
| 38 | 94.6 | |
| 39 | 94.6 | |
| 40 | 94.5 | |
| 41 | 94.5 | |
| 42 | 94.5 | |
| 43 | 94.4 | |
| 44 | 94.3 | |
| 45 | 94.2 | |
| 46 | 94.2 | |
| 47 | 94.1 | |
| 48 | 94.1 | |
| 49 | 94.1 | |
| 50 | 94.0 |