TAU2-bench leaderboard
43 models ranked, highest score first.
| # | Model | Company | Score |
|---|---|---|---|
| 1 | JT-35B-Flash | China Mobile | 99.1% |
| 2 | Claude Fable 5 | Anthropic | 98.5% |
| 3 | Step 3.7 Flash | StepFun | 98.5% |
| 4 | Grok 4.3 | xAI | 97.7% |
| 5 | DeepSeek V4 Pro | DeepSeek | 96.2% |
| 6 | Qwen3.6 Max Preview | Alibaba | 95.9% |
| 7 | DeepSeek V4 Flash | DeepSeek | 95.0% |
| 8 | Qwen3.7 Max | Alibaba | 94.7% |
| 9 | Claude Opus 4.8 | Anthropic | 94.4% |
| 10 | Mistral Medium 3.5 | Mistral | 94.2% |
| 11 | Qwen3.6 27B | Alibaba | 94.2% |
| 12 | MiMo-V2.5-Pro | Xiaomi | 94.2% |
| 13 | GPT-5.5 | OpenAI | 93.9% |
| 14 | Qwen3.7 Plus | Alibaba | 93.0% |
| 15 | Hy3-preview | Tencent | 92.7% |
| 16 | Ring-2.6-1T | InclusionAI | 92.4% |
| 17 | MiMo-V2.5 | Xiaomi | 90.6% |
| 18 | DeepSeek V3.2 | DeepSeek | 90.6% |
| 19 | Ling-2.6-1T | InclusionAI | 89.8% |
| 20 | Claude Opus 4.5 | Anthropic | 89.5% |
| 21 | MiniMax-M3 | MiniMax | 88.9% |
| 22 | MiniCPM-V 4.6 1.3B | OpenBMB | 87.7% |
| 23 | Ling 2.6 Flash | InclusionAI | 86.0% |
| 24 | GPT-5.3 Codex | OpenAI | 86.0% |
| 25 | GPT-5.2 | OpenAI | 84.8% |
| 26 | GPT-5.1 | OpenAI | 81.9% |
| 27 | MiniCPM5-1B | OpenBMB | 81.0% |
| 28 | Claude Sonnet 4.6 | Anthropic | 78.9% |
| 29 | GPT-5 | OpenAI | 67.0% |
| 30 | HyperNova 60B 2605 | Multiverse Computing | 63.2% |
| 31 | Gemini 3.5 Flash | 58.8% | |
| 32 | GPT-5.5 Instant | OpenAI | 49.4% |
| 33 | Granite 4.1 30B | IBM | 42.1% |
| 34 | North Mini Code | Cohere | 37.4% |
| 35 | Gemma 4 12B | 36.3% | |
| 36 | Gemini 2.5 Flash | 31.6% | |
| 37 | Gemini 2.0 Flash | 29.5% | |
| 38 | Granite 4.1 8B | IBM | 27.8% |
| 39 | Mistral Large 3 | Mistral | 24.6% |
| 40 | DeepSeek-V3 | DeepSeek | 22.8% |
| 41 | Claude 3 Haiku | Anthropic | 21.1% |
| 42 | Granite 4.1 3B | IBM | 19.6% |
| 43 | LFM2.5-8B-A1B | Liquid AI | 16.1% |