List your company

Safety Alignment Evaluation on StrongReject SR-PAPA

100Safety RateMESA
67.104875.644984.18592.7251May 30, 2026

Evaluation results

MethodLinks
MESAArchitecture=DeepSeek-v2-Lite2026.05100
GRPOArchitecture=Qwen3-30B-A3B2026.0599.68
Stair-SFTArchitecture=Qwen3-30B-A3B2026.0599.68
Stair-DPOArchitecture=Qwen3-30B-A3B2026.0599.68
MESAArchitecture=Qwen3-30B-A3B2026.0599.68
SafeXArchitecture=Qwen3-30B-A3B2026.0598.72
SFTArchitecture=Qwen3-30B-A3B2026.0597.76
Stair-DPOArchitecture=DeepSeek-v2-Lite2026.0596.17
Stair-SFTArchitecture=DeepSeek-v2-Lite2026.0593.29
SFTArchitecture=DeepSeek-v2-Lite2026.0591.05
Base(instruct)Architecture=Qwen3-30B-A3B2026.0591.05
SafeXArchitecture=DeepSeek-v2-Lite2026.0587.54
Base(chat)Architecture=DeepSeek-v2-Lite2026.0569.01
GRPOArchitecture=DeepSeek-v2-Lite2026.0568.37