LLM Safety and Informativeness Evaluation on HarmBench
82.5Safety RateSafeMoE-XL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SafeMoE-XLVariant=XL2026.05 | 82.5 | 7.5 | |
| Oyster-I 14BParameters=14B2026.05 | 75.4 | 7.1 | |
| SafeMoE-QwenBackbone=Qwen2026.05 | 71.9 | 7.5 | |
| RealSafe-R12026.05 | 59.3 | 6.7 | |
| Zephyr2026.05 | 52.4 | 6.3 | |
| SN-Tune2026.05 | 49.6 | 6.1 | |
| Deepseek-R1-QwenBackbone=Qwen2026.05 | 49.1 | 6.5 | |
| Mistral-SFT2026.05 | 30.3 | 6.6 | |
| Qwen-3BParameters=3B2026.05 | 27.5 | 4.1 | |
| SafeLoRA2026.05 | 20.7 | 6.3 | |
| Mistral-7BParameters=7B2026.05 | 14.3 | 3.5 |