Senior SWE-Bench Basic Solve Rate
Code · 2026-06-30
Senior SWE-Bench basic solve rate counts scored tasks whose functional requirements are satisfied without the stricter taste and engineering-quality criteria. The 100-task suite's displayed leaderboard has model-specific scored coverage; score notes record each run's denominator, and no common denominator is assumed.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5 | 62.1 |
| Claude Opus 5.5 | 61.8 |
| GPT-6 Astra | 60.0 |
| Claude Fable 5.1 | 57.9 |
| Claude Fable 5 | 53.7 |
| GPT-5.6 Sol | 53.7 |
| GPT-5.5 | 52.6 |
| Grok 4.6 | 51.6 |
| Sonnet 5 | 50.5 |
| Grok 4.5 | 50.5 |
| Grok 4.7 | 49.5 |
| Opus 4.7 | 46.8 |
| Opus 4.8 | 44.2 |
| Gemini 3.7 Flash | 44.2 |
| DeepSeek V4.1 Flash | 44.0 |
| GPT-5.4 | 41.1 |
| MiniMax M3 | 41.1 |
| Kimi K3 | 40.4 |
| GPT-5.6 Terra | 36.8 |
| GLM-5.2 | 35.8 |
| Gemini 3.5 Flash | 22.1 |
| Gemini 3.1 Pro Preview | 9.5 |