tau2-bench — Sierra tau2-bench, retail domain; Kimi K2 card
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Claude Opus 4.6 | Anthropic | 91.9 | 1 | 2026-08-23 |
| 2 | Claude Sonnet 4.6 | Anthropic | 91.7 | 1 | 2026-08-23 |
| 3 | Claude Opus 4.5 | Anthropic | 88.9 | 1 | 2026-08-23 |
| 4 | Claude Haiku 4.5 | Anthropic | 83.2 | 1 | 2026-08-23 |
| 5 | GPT-5.2 | OpenAI | 82 | 1 | 2026-08-23 |
| 6 | Claude 4 Opus | Anthropic | 81.8 | 2 | 2026-06-15 |
| 7 | GPT-5 | OpenAI | 81.1 | 1 | 2026-08-23 |
| 8 | o3 | OpenAI | 80.2 | 1 | 2026-08-23 |
| 9 | GPT-5.1 Instant | OpenAI | 77.9 | 1 | 2026-08-23 |
| 10 | GPT-5.1 | OpenAI | 77.9 | 1 | 2026-08-23 |
| 11 | Claude Sonnet 4 | Anthropic | 75 | 2 | 2026-06-15 |
| 12 | LongCat Flash Lite | Meituan | 73.1 | 1 | 2026-08-23 |
| 13 | Qwen3.5 4B | Alibaba | 71.9 | 1 | 2026-08-09 |
| 14 | Qwen3 235B A22B Instruct 2507 | Alibaba | 71.3 | 1 | 2026-08-23 |
| 15 | Kimi K2 Instruct | Moonshot | 70.6 | 4 | 2026-08-23 |
| 16 | DeepSeek-V3 | DeepSeek | 69.1 | 2 | 2026-06-15 |
| 17 | Qwen3 Next 80B A3B | Alibaba | 67.8 | 1 | 2026-08-23 |
| 18 | GPT-4o | OpenAI | 63.4 | 1 | 2026-08-23 |
| 19 | Nemotron 3 Super 120B A12B | NVIDIA | 62.8 | 1 | 2026-08-23 |
| 20 | Qwen3 Next 80B A3B Instruct | Alibaba | 57.3 | 1 | 2026-08-23 |
| 21 | Qwen3 235B A22B | Alibaba | 57 | 2 | 2026-06-15 |
| 22 | NVIDIA Nemotron 3 Nano 30B A3B | NVIDIA | 56.9 | 1 | 2026-08-23 |
| 23 | Qwen3 30B A3B | Alibaba | 56.1 | 1 | 2026-08-09 |
| 24 | Gemma 4 E4B | 42.1 | 1 | 2026-08-09 | |
| 25 | LFM2.5-8B-A1B | Liquid AI | 39.8 | 1 | 2026-08-09 |
| 26 | Gemma 4 E2B | 18.9 | 1 | 2026-08-09 | |
| 27 | Granite 4.0 H Tiny | IBM | 18.4 | 1 | 2026-08-09 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.