ResearchBenchmarksPointGoal on Safety GymFollow0.64Normalized RewardReference policy (πref)0.20320.31660.430.5434May 20, 2026Evaluation ResultsMethodMethodLinksNormalized RewardNormalized CostReference policy (πref)2026.050.642.82PPL2026.050.460.95PREFINE2026.050.40.24CPQ2026.050.250.2SafeDICE2026.050.240.26BC2026.050.220.82