| Rank | Model | Provider | Score |
|---|---|---|---|
| 1 | DeepSeek V4.1 Flash | DeepSeek | 74.2% |
| 2 | GPT 6 Astra | OpenAI | 74.0% |
| 3 | Claude Opus 5 | Anthropic | 74.0% |
| 4 | GPT 5.6 Sol | OpenAI | 73.0% |
| 5 | MiMo V2.6 Pro | Xiaomi | 71.9% |
| 6 | Claude Fable 5 | Anthropic | 70.0% |
| 7 | MiMo V2.6 Flash | Xiaomi | 67.9% |
| 8 | Kimi K3 | Moonshot AI | 67.5% |
| 9 | GLM 5.3 | Zhipu AI | 66.9% |
| 10 | GLM 5.3 Flash | Zhipu AI | 63.4% |
| 11 | DeepSeek V4 Pro 0813 | DeepSeek | 62.7% |
| 12 | Qwen3.8 Flash Next | Alibaba | 58.7% |
| 13 | Qwen3.8 Max | Alibaba | 56.6% |
| 14 | DeepSeek V4 Flash 0731 | DeepSeek | 54.4% |
| 15 | Qwen3.8 27B | Alibaba | 42.2% |
| 16 | MiMo V2.5 Pro | Xiaomi | 19.0% |
| 17 | Qwen3.6 27B | Alibaba | 13.3% |
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.