SWE-bench Verified (Vals Index 102-task subset)
Code · 2026-07-01
SWE-bench Verified evaluated on the 102-task Vals Index subset: 33 sampled tasks from each of the first three human-duration bands plus all three tasks in the hardest band. It uses the Vals mini-swe-agent bash harness and reports issue-resolution rate.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-5.6 Sol | 99.0 |
| Claude Fable 5 | 97.1 |
| Claude Opus 5 | 95.1 |
| Kimi K3 | 95.1 |
| GPT-5.6 Luna | 93.1 |
| Opus 4.8 | 89.2 |
| Grok 4.5 | 86.3 |
| GLM-5.2 | 83.3 |
| GPT-5.5 | 82.6 |
| Opus 4.7 | 82.4 |
| Muse Spark 1.1 | 78.4 |
| Sonnet 4.6 | 77.5 |
| Gemini 3.6 Flash | 77.5 |
| Sonnet 5 | 75.5 |
| DeepSeek-V4-Pro | 75.5 |
| Gemini 3.5 Flash | 75.5 |
| Inkling | 75.5 |
| Inkling-Small | 74.5 |
| Kimi K2.6 | 74.5 |
| GPT-5.6 Terra | 73.5 |
| Gemini 3.1 Pro Preview | 72.5 |
| GLM-5.1 | 72.5 |
| Grok 4.3 | 70.6 |
| Qwen3.6 Plus (2026-04-02) | 70.6 |
| MiniMax M2.7 | 69.6 |
| MiniMax M3 | 69.6 |
| Gemini 3.5 Flash-Lite | 68.6 |
| Gemini 3 Flash Preview | 68.6 |
| GPT-5.4 Mini | 67.6 |
| GPT-5.4 Nano | 67.6 |
| Nemotron 3 Ultra 550B A55B | 67.6 |
| Grok 4.20 | 66.7 |
| MiMo-V2.5-Pro | 66.7 |
| Qwen3.7-Max | 66.7 |
| Qwen3.7-Plus | 66.7 |
| Haiku 4.5 | 64.7 |
| MiMo-V2.5 | 64.7 |
| Mistral Medium 3.5 | 64.7 |
| Gemini 3.1 Flash-Lite Preview | 57.8 |
| Laguna M.1 | 51.0 |