tau-bench

tau-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains

↓ tau2-bench

零售(115 个任务)和航空(50 个任务)领域的模拟客服对话,智能体需与 LLM 模拟的用户交谈、调用数据库 API 并遵守书面政策。仅当最终数据库状态与标注目标一致时任务才算通过;pass^k 衡量同一任务在全部 k 次独立试验中是否都成功,从而暴露智能体的不一致性。是遵循政策的工具智能体的参考基准。

69.2%
TC (claude-3-5-sonnet-20241022), retail
官方榜单
2024-06-17 → 2024-10-22: 35.2% → 69.2%
发布
2024-06
维护者
Sierra Research (Yao et al.)
状态
retired
污染风险
high
指标
pass^1 (percent, ↑)
题量
165
领域
tool-use instruction-following general-assistant
人类
无实测基线

备注. The repository README now warns that its airline and retail tasks are outdated and directs users to the tau2-bench repository (tau^3-bench) for fixed tasks; developer reports of 'tau-bench' after mid-2025 usually mean tau2-bench. Scores depend on the user-simulator model, which is not standardised.

完整账本

系统开发者分数日期来源条件
TC (claude-3-5-sonnet-20241022), retailAnthropic69.2%官方榜单split: retail scaffold: function calling (TC)
Repository README leaderboard; date is the model's release date (README gives none).
gpt-4o (function calling), tau-retailOpenAI61.2%论文split: retail scaffold: function calling (TC)
Table 2 of the paper; user simulated by gpt-4.
TC (claude-3-5-sonnet-20241022), airlineAnthropic46%官方榜单split: airline scaffold: function calling (TC)
Repository README leaderboard; date is the model's release date (README gives none).
gpt-4o (function calling), tau-airlineOpenAI35.2%论文split: airline scaffold: function calling (TC)
Table 2 of the paper; user simulated by gpt-4.