LLM Safety and Informativeness Evaluation on BeaverTails
87.2Safety RateSafeMoE-XL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SafeMoE-XLVariant=XL2026.05 | 87.2 | 6.4 | |
| Oyster-I 14BParameters=14B2026.05 | 81.7 | 7.6 | |
| RealSafe-R12026.05 | 68.8 | 5.6 | |
| SafeMoE-QwenBackbone=Qwen2026.05 | 63.4 | 6.9 | |
| Deepseek-R1-QwenBackbone=Qwen2026.05 | 53.6 | 6.8 | |
| SN-Tune2026.05 | 51.6 | 5.2 | |
| Zephyr2026.05 | 39.8 | 5.9 | |
| Qwen-3BParameters=3B2026.05 | 34 | 6.2 | |
| Mistral-7BParameters=7B2026.05 | 31.4 | 3.9 | |
| Mistral-SFT2026.05 | 22.4 | 5.9 | |
| SafeLoRA2026.05 | 15.7 | 4.5 |