Snorkel SWE-bench CLI pass@5
Code
Published pass@5 on the 200-task SWE-bench CLI release; a correlated diagnostic of the pass@1 leaderboard.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 29.7 |
| Claude Fable 5.1 | 26.2 |
| Claude Opus 5 | 24.7 |
| GPT-6 Astra | 18.8 |
| Grok 4.7 | 17.1 |
| Gemini 3.8 Flash | 14.1 |
| Grok 4.6 | 13.0 |
| Kimi K3 | 11.9 |
| GLM-5.3 | 10.8 |
| Muse Spark 1.3 | 5.7 |