ResearchBenchmarksAntVelocity on Safety GymFollow0.93Normalized RewardReference policy (πref)0.21240.39870.5850.7713May 20, 2026Evaluation ResultsMethodMethodLinksNormalized RewardNormalized CostReference policy (πref)2026.050.938.04PREFINE2026.050.930.13PPL2026.050.90.66BC2026.050.814.23SafeDICE2026.050.790.5CPQ2026.050.240.1