List your company

User Simulation Behavioral Alignment on tau2-bench Retail + Airline (test)

95.8HL ScoreHumans
8.9631.50554.0576.595May 13, 2026

Evaluation results

MethodLinks
HumansUser Simulator=DeepSeek-V3.12026.0595.862.37994.997.888.692.293.4
PPol: EvolvedUser Simulator=DeepSeek-V3.12026.055765.761.465.493.78454.874.5
PPol: InitialUser Simulator=DeepSeek-V3.12026.054132.336.742.785.343.833.251.2
DP PersonasUser Simulator=DeepSeek-V3.12026.0529.23431.637.175.949.832.948.9
Base-simulatorUser Simulator=DeepSeek-V3.12026.0512.314.613.53288.753.454.457.1