| Rank | Model | Provider | Score |
|---|---|---|---|
| 1 | o1 preview | OpenAI | 90.8% |
| 2 | Claude 3.5 Sonnet | Anthropic | 88.7% |
| 3 | GPT 4o | OpenAI | 88.7% |
| 4 | Llama 3.1 405B | Meta | 87.3% |
| 5 | Claude 3 Opus | Anthropic | 86.8% |
| 6 | Qwen2.5 72B | Alibaba | 86.1% |
| 7 | GLM 4 Plus | Zhipu AI | 86.0% |
| 8 | Gemini 1.5 Pro | 85.9% | |
| 9 | o1 mini | OpenAI | 85.2% |
| 10 | Yi Large | 01.AI | 84.6% |
| 11 | Mistral Large 2 | Mistral AI | 84.0% |
| 12 | Llama 3.1 70B | Meta | 83.6% |
| 13 | GPT 4o mini | OpenAI | 82.0% |
| 14 | DeepSeek V2.5 | DeepSeek | 80.6% |
| 15 | Gemini 1.5 Flash | 78.9% | |
| 16 | Llama 3.1 8B | Meta | 69.4% |
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.