HumanEval
The HumanEval dataset contains 164 hand-written programming problems.
State-of-the-Art Leaderboard
| Rank | Model | Metric | Score | Paper | Date |
|---|---|---|---|---|---|
| 🏆 1 | Claude Opus 5.5 | Pass@1 | 92.0% | - | 2026-09-10 |
| 2 | GPT-6 Astra | Pass@1 | 90.1% | GPT-4 Technical Report | 2026-09-01 |