List your company

Safety Alignment Evaluation on StrongReject SR-PAPL

100Safety RateMESA
78.399284.007189.61595.2229May 30, 2026

Evaluation results

MethodLinks
MESAArchitecture=DeepSeek-v2-Lite2026.05100
Stair-DPOArchitecture=Qwen3-30B-A3B2026.0599.68
MESAArchitecture=Qwen3-30B-A3B2026.0599.68
Stair-DPOArchitecture=DeepSeek-v2-Lite2026.0599.36
SFTArchitecture=Qwen3-30B-A3B2026.0599.36
Stair-SFTArchitecture=Qwen3-30B-A3B2026.0599.04
SafeXArchitecture=Qwen3-30B-A3B2026.0599.04
Stair-SFTArchitecture=DeepSeek-v2-Lite2026.0598.08
GRPOArchitecture=Qwen3-30B-A3B2026.0598.08
SFTArchitecture=DeepSeek-v2-Lite2026.0595.53
SafeXArchitecture=DeepSeek-v2-Lite2026.0590.74
Base(instruct)Architecture=Qwen3-30B-A3B2026.0583.71
GRPOArchitecture=DeepSeek-v2-Lite2026.0579.55
Base(chat)Architecture=DeepSeek-v2-Lite2026.0579.23