benchmark · first seen 8 Oct 2026
tau-bench
Also: τ-bench, τ2-bench, tau2-bench, tau-bench benchmark, dual-control benchmark, tool-calling benchmark
A benchmark evaluating LLM agents in tool-use customer service tasks with reward-based scoring.
Mentions per week
PapersCodeDiscussion
Earliest sources
- TRACE: Diagnosing Verifier Brittleness in Agentic EvaluationPapers · 8 Oct 2026 · Radhika Gaonkar
- Program Learning with Verifiable Rewards: Distilling Expert Traces into ProgramsPapers · 13 Oct 2026 · Vishvesh Bhat, Muhammad Ahmed Mohsin, Asad Aali
- J-tau: A Fully Localized Japanese Benchmark for Tool-Agent-User InteractionPapers · 14 Oct 2026 · Chihiro Yano, Jun Hirako, Hiroaki Yamagiwa, Ryota Hirobuchi
- Does Multi-Agent Collaboration Pay When Agents Serve Users? Testing Agent Scaling in Dual-Control SettingsPapers · 20 Oct 2026 · Ethan Yang, Xiangcheng Shi
- GRADUS: Discovering and Recovering Agent Lessons from the Environment's Latent SyllabusPapers · 20 Oct 2026 · Qing Ping, Wei Yang, Panpan Xu, Johannes Kirmayr, Lukas Stappen, Junjie Tang, Ju…
- Concordat: Typed Sessions for LLM AgentsPapers · 27 Oct 2026 · Milad Sikaroudi
Latest sources
- When Evaluators of AI Agents Disagree: An Empirical Study of Native Scoring, LLM Judges, and Commit-Boundary Checks on tau2-benchPapers · 10 Nov 2026 · Sungbin Im
- Knowing When to Intervene: Causal Evaluation of AI AgentsPapers · 6 Nov 2026 · Ayush Sawarni, Wenqian Xing, Tijana Zrnic, Ramesh Johari, Vasilis Syrgkanis
- Think Before You Act: Decision Gate for LLM AgentsPapers · 28 Oct 2026 · Iftach Shoham, Yuval David, Offer Akrabi, Erez Bilgory, Ido Levy, Sami Marreed,…
- Concordat: Typed Sessions for LLM AgentsPapers · 27 Oct 2026 · Milad Sikaroudi
- GRADUS: Discovering and Recovering Agent Lessons from the Environment's Latent SyllabusPapers · 20 Oct 2026 · Qing Ping, Wei Yang, Panpan Xu, Johannes Kirmayr, Lukas Stappen, Junjie Tang, Ju…
- Does Multi-Agent Collaboration Pay When Agents Serve Users? Testing Agent Scaling in Dual-Control SettingsPapers · 20 Oct 2026 · Ethan Yang, Xiangcheng Shi
- J-tau: A Fully Localized Japanese Benchmark for Tool-Agent-User InteractionPapers · 14 Oct 2026 · Chihiro Yano, Jun Hirako, Hiroaki Yamagiwa, Ryota Hirobuchi
- Program Learning with Verifiable Rewards: Distilling Expert Traces into ProgramsPapers · 13 Oct 2026 · Vishvesh Bhat, Muhammad Ahmed Mohsin, Asad Aali
- TRACE: Diagnosing Verifier Brittleness in Agentic EvaluationPapers · 8 Oct 2026 · Radhika Gaonkar