Benchmark
TAU-bench Retail
reasoning
communication
text
A benchmark for evaluating tool-agent-user interaction in retail environments. Tests language agents' ability to handle dynamic conversations with users while using domain-specific API tools and following policy guidelines. Evaluates agents on tasks like order cancellations, address changes, and order status checks through multi-turn conversations.
语言EN
满分1
参评模型22
模型排名
| 名次 | 模型 | 机构 | 分数 | 来源 |
|---|---|---|---|---|
| 1 | Claude Sonnet 4.5 | Anthropic | 86.2 | 来源 ↗ |
| 2 | Claude Opus 4.1 | Anthropic | 82.4 | 来源 ↗ |
| 3 | Claude Opus 4 | Anthropic | 81.4 | 来源 ↗ |
| 4 | Claude 3.7 Sonnet | Anthropic | 81.2 | 来源 ↗ |
| 5 | Claude Sonnet 4 | Anthropic | 80.5 | 来源 ↗ |
| 6 | GLM-4.5 | Zhipu AI | 79.7 | 来源 ↗ |
| 7 | GLM-4.5-Air | Zhipu AI | 77.9 | 来源 ↗ |
| 8 | o4-mini | OpenAI | 71.8 | 来源 ↗ |
| 9 | o1 | OpenAI | 70.8 | 来源 ↗ |
| 10 | Qwen3-Next-80B-A3B-Thinking | Alibaba Cloud / Qwen Team | 69.6 | 来源 ↗ |
| 11 | Claude 3.5 Sonnet | Anthropic | 69.2 | 来源 ↗ |
| 12 | GPT-4.5 | OpenAI | 68.4 | 来源 ↗ |
| 13 | GPT-4.1 | OpenAI | 68.0 | 来源 ↗ |
| 14 | Qwen3-235B-A22B-Thinking-2507 | Alibaba Cloud / Qwen Team | 67.8 | 来源 ↗ |
| 15 | GPT OSS 120B | OpenAI | 67.8 | 来源 ↗ |
| 16 | Qwen3-Next-80B-A3B-Instruct | Alibaba Cloud / Qwen Team | 60.9 | 来源 ↗ |
| 17 | GPT-4o | OpenAI | 60.3 | 来源 ↗ |
| 18 | o3-mini | OpenAI | 57.6 | 来源 ↗ |
| 19 | GPT-4.1 mini | OpenAI | 55.8 | 来源 ↗ |
| 20 | GPT OSS 20B | OpenAI | 54.8 | 来源 ↗ |
| 21 | Claude 3.5 Haiku | Anthropic | 51.0 | 来源 ↗ |
| 22 | GPT-4.1 nano | OpenAI | 22.6 | 来源 ↗ |