ResearchBenchmarksAgentic Task Completion on τ2-bench RetailFollow100Success RateRAC89.692.39597.7May 5, 2026Evaluation ResultsMethodMethodLinksSuccess RateCompletion RateTime (s)Token Usage (k)RACFramework=RACFramework=RAC2026.05100892266,000SagaLLMFramework=SagaLLMFramework=SagaLLM2026.0598—6571,000LangGraphFramework=LG(PE), Prom...Framework=LG(PE), Prompt Engineering=true2026.0591—3052,000LangGraphFramework=LGFramework=LG2026.0590—2849,000