ResearchBenchmarksHarmfulness Evaluation on HEx-PHI, AdvBench, HarmBench, & JailbreakBench FullFollow4.95Average Score (Full)RLVR0.98762.01633.0454.0737Oct 17, 2025Evaluation ResultsMethodMethodLinksAverage Score (Full)Score DifferenceRLVR2025.104.95-0.01RLVR2025.104.95-0.01SFT2025.104.78-0.03SFT2025.104.78-0.03Base2025.101.14-0.01Base2025.101.14-0.01