HumanEvalFix
humanevalfix@1.0HumanEval functions with a planted bug; the agent repairs each one until its tests pass. Its free boards run on a local model, so they need no API key.
How HumanEvalFix measures up to the dataset checklist
Boards
Every board here runs on an allowlisted network.
| Model | Highest pass rate | Attempts | Harnesses | Updated |
|---|---|---|---|---|
| Qwen2.5-Coder 7Bsample-40 task set, 40 tasks | 1 | 1 | 18 Oct 2026 |