Safety Alignment Evaluation on Strata
99Safety RateMESA
Evaluation Results
| Method | Links | |
|---|---|---|
| MESAArchitecture=Qwen3-30B-A3B2026.05 | 99 | |
| GRPOArchitecture=Qwen3-30B-A3B2026.05 | 96 | |
| MESAArchitecture=DeepSeek-v2-Lite2026.05 | 95 | |
| Stair-SFTArchitecture=Qwen3-30B-A3B2026.05 | 94 | |
| Stair-DPOArchitecture=DeepSeek-v2-Lite2026.05 | 93 | |
| Stair-SFTArchitecture=DeepSeek-v2-Lite2026.05 | 92.5 | |
| SFTArchitecture=DeepSeek-v2-Lite2026.05 | 92 | |
| SafeXArchitecture=Qwen3-30B-A3B2026.05 | 91 | |
| SFTArchitecture=Qwen3-30B-A3B2026.05 | 88.5 | |
| Base(instruct)Architecture=Qwen3-30B-A3B2026.05 | 88 | |
| Stair-DPOArchitecture=Qwen3-30B-A3B2026.05 | 87.5 | |
| SafeXArchitecture=DeepSeek-v2-Lite2026.05 | 81 | |
| Base(chat)Architecture=DeepSeek-v2-Lite2026.05 | 70.5 | |
| GRPOArchitecture=DeepSeek-v2-Lite2026.05 | 64 |