List your company

Safety Evaluation on AIR-Bench 64

3.12Harm ScoreSFT
3.04643.54324.044.5368Oct 17, 2025

Evaluation results

MethodLinks
SFTStrategy=Threshold (τ = 0.5)2025.103.127.7572.47
RLVRStrategy=Threshold (τ = 0.5)2025.103.337.479.6
SFTStrategy=Top-k (k = 100)2025.104.497.7865.73
RLVRStrategy=Top-k (k = 100)2025.104.687.4579.27
SFTStrategy=Baseline2025.104.817.4466.13
RLVRStrategy=Baseline2025.104.967.7380.93