tau2-bench

A new study of 9,876 trajectories from eight frontier model families found that agents falsely claim completion in 45–48% of failures on tau2-bench and 75.8% on AppWorld, while LLM judge models topped…

A new framework called WER (Write, Execute, and Refine) trains a Skill Optimizer to improve tool-using agents' performance by learning from execution feedback, achieving a 7.80-point average Pass@1 im…

A new arXiv study (2608.14641v1) evaluating four open-source model routers across RouterBench, BFCL v4, tau2-bench, and WebArena found that only vLLM Semantic Router varies with prompt content, yet it…