TAU-bench

The 'TAU-bench' dataset/benchmark contains a collection of tasks designed to evaluate AI agents' performance in multi-turn and multi-step interactions involving the use of external tools.

Papers using TAU-bench (11)

Self-Challenging Language Model Agents2025 · 1 citesSAMULE: Self-Learning Agents Enhanced by Multi-level Reflection2025 · 6 citesSkill or Skip? Learning Selective Skill Invocation in Agentic Tasks via Dual-Granularity Preference Learning2026RealUserSim: Bridging the Reality Gap in Agent Benchmarking via Grounded User Simulation2026MAVEN: Improving Generalization in Agentic Tool Calling2026Robust Tool Use via Fission-GRPO: Learning to Recover from Execution Errors2026CM2: Reinforcement Learning with Checklist Rewards for Multi-Turn and Multi-Step Agentic Tool Use2026Prompting Policies for Multi-step Reasoning and Tool-Use in Black-box LLMs with Iterative Distillation of Experience2026Multi-Turn Reinforcement Learning for Tool-Calling Agents with Iterative Reward Calibration2026On Generalization in Agentic Tool Calling: CoreThink Agentic Reasoner and MAVEN Dataset2025Establishing Best Practices For Building Rigorous Agentic Benchmarks2025