List your company

Safety Alignment Evaluation on StrongReject SR-Pair

98.72Safety RateStair-DPO
50.214462.807275.487.9928May 30, 2026

Evaluation results

MethodLinks
Stair-DPOArchitecture=Qwen3-30B-A3B2026.0598.72
Stair-SFTArchitecture=Qwen3-30B-A3B2026.0596.59
GRPOArchitecture=Qwen3-30B-A3B2026.0591.37
MESAArchitecture=Qwen3-30B-A3B2026.0590.73
SFTArchitecture=Qwen3-30B-A3B2026.0587.22
SafeXArchitecture=Qwen3-30B-A3B2026.0586.58
Stair-DPOArchitecture=DeepSeek-v2-Lite2026.0576.36
SFTArchitecture=DeepSeek-v2-Lite2026.0575.08
MESAArchitecture=DeepSeek-v2-Lite2026.0573.48
Stair-SFTArchitecture=DeepSeek-v2-Lite2026.0572.03
Base(instruct)Architecture=Qwen3-30B-A3B2026.0566.77
SafeXArchitecture=DeepSeek-v2-Lite2026.0556.87
GRPOArchitecture=DeepSeek-v2-Lite2026.0555.27
Base(chat)Architecture=DeepSeek-v2-Lite2026.0552.08