List your company

Agent Task Performance on τ-bench retail 20-task (held-out)

0.519GOffline AW Controller
0.329720.378860.4280.47714Jul 5, 2026

Evaluation results

MethodLinks
Offline AW ControllerHarness configuration=AW, Scoring Protocol=Adapted Harness MDP2026.070.51917.2
Base HarnessHarness configuration=Base, Scoring Protocol=Adapted Harness MDP2026.070.337—