SWE-rebench pass@5
Code · 2025-05-14
SWE-rebench pass@5 reports the percentage of software-engineering tasks solved by at least one of five sampled attempts under the benchmark's evaluation setup.
Top models (higher is better)
| Model | Score |
|---|---|
| MiniMax M3 | 69.4 |
| MiMo-V2.5-Pro | 65.8 |
| Qwen3.6 27B | 57.7 |
| Qwen3.6 35B A3B | 43.2 |