| Rank | Model | Provider | Score |
|---|---|---|---|
| 1 | Claude Opus 5 | Anthropic | 75.3% |
| 2 | DeepSeek V4.1 Flash | DeepSeek | 65.4% |
| 3 | DeepSeek V4 Pro 0813 | DeepSeek | 61.5% |
| 4 | GLM 5.3 | Zhipu AI | 58.0% |
| 5 | Kimi K3 | Moonshot AI | 58.0% |
| 6 | GPT 5.6 Sol | OpenAI | 56.8% |
| 7 | GLM 5.3 Flash | Zhipu AI | 56.3% |
| 8 | Qwen3.8 Max | Alibaba | 55.9% |
| 9 | DeepSeek V4 Flash 0731 | DeepSeek | 54.2% |
| 10 | Qwen3.8 Flash Next | Alibaba | 48.1% |
| 11 | Qwen3.8 27B | Alibaba | 42.3% |
| 12 | Qwen3.6 Plus | Alibaba | 37.9% |
| 13 | Qwen3.6 27B | Alibaba | 36.2% |
| 14 | Qwen3.6 35B A3B | Alibaba | 29.4% |
Figures are taken from public model cards, technical reports and vendor announcements, and are approximate where sources disagree. Results from different benchmark versions or tool-use settings never share a column: Terminal Bench 2.1, 3.0 and DeepSWE v1.1 are distinct benchmarks, and HLE with tools is separate from HLE without tools. Only models with a published result are listed; the score column is shaded relative to the best and worst result on this benchmark.