List your company

Harmfulness Evaluation on HarmBench

1.12Harmful Response RatioBase
-1.526416.336834.252.0632Oct 17, 2025Nov 3, 2025Nov 21, 2025Dec 8, 2025Dec 26, 2025Jan 12, 2026Jan 30, 2026

Evaluation results

MethodLinks
Base2025.101.12
Base2025.101.12
SFT2025.104.79
SFT2025.104.79
RLVR2025.104.96
RLVR2025.104.96
STAR-1Model Size=8B2026.0121.26
THINKSAFEModel Size=8B2026.0127.08
DirectRefusalModel Size=8B2026.0132.39
SafeKeyModel Size=8B2026.0132.72
THINKSAFEModel Size=7B2026.0140.2
SafeChainModel Size=8B2026.0144.52
SafeKeyModel Size=7B2026.0145.35
SafePathModel Size=8B2026.0147.51
SafeChainModel Size=7B2026.0151
DirectRefusalModel Size=7B2026.0152.33
InitialModel Size=8B2026.0152.33
THINKSAFEModel Size=1.5B2026.0152.99
STAR-1Model Size=7B2026.0152.99
SafePathModel Size=7B2026.0155.15
InitialModel Size=7B2026.0156.98
SafeKeyModel Size=1.5B2026.0158.8
SafeChainModel Size=1.5B2026.0159.3
STAR-1Model Size=1.5B2026.0162.79
SafePathModel Size=1.5B2026.0165.28
DirectRefusalModel Size=1.5B2026.0166.11
InitialModel Size=1.5B2026.0167.28