List your company

Safety Guardrail on Response-side safety benchmarks

87F1 ScoreStage3-DPO
82.21683.45884.785.942Jul 7, 2026

Evaluation results

MethodLinks
Stage3-DPOParams=4B2026.0787
Qwen3Guard-8B-GenParams=8B2026.0786.3
Stage2-SFTParams=4B2026.0786.3
Stage1-SFT-v4Params=4B2026.0786
Qwen3Guard-0.8B-GenParams=0.8B2026.0785.8
Qwen3Guard-4B-GenParams=4B2026.0785.3
Stage1-SFT-v1Params=4B2026.0784.8
Stage1-SFT-v0Params=4B2026.0784.7
YuFeng-XGuard-Reason-8BParams=8B2026.0784.1
Stage1-SFT-v3Params=4B2026.0783.9
YuFeng-XGuard-Reason-0.6BParams=0.6B2026.0783.5
Stage1-SFT-v2Params=4B2026.0782.4