Benchmark
End-to-end tasks completed in a real terminal environment (v3.0 task set).
Score history →| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 5 | 43.3% |
| 2 | GPT 5.6-Sol | 34.4% |
| 3 | DeepSeek V4.1-Flash 0910 | 30% |
| 4 | GLM 5.3 | 28.3% |
| 5 | Kimi K3 | 17.7% |
| 6 | DeepSeek V4-Pro 0813 | 11.8% |
| 7 | DeepSeek V4-Flash 0731 | 7.6% |