Safety Evaluation on OrBench Toxic
81.7Safety ScoreMistral MoE-XL
Evaluation Results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Mistral MoE-XLBackbone=Mistral-7B, Method=SafeMoE2026.05 | 81.7 | 6.76 | 75 | 90 | 80 | 90 | 85 | 70 | 80 | 75 | 82 | 90 | 81.7 | |
| Qwen MoE-XLBackbone=Qwen-3B, Method=SafeMoE2026.05 | 54.89 | 7.25 | 42.5 | 64.3 | 44.8 | 61.4 | 45.8 | 51 | 76.1 | 69 | 51 | 43 | 54.89 | |
| Qwen-3BBackbone=Qwen-3B, Training=Standard instruction-tuned2026.05 | 20.15 | 6.62 | 6 | 5.9 | 13.6 | 40 | 13.5 | 6 | 25.6 | 63 | 6.7 | 20.5 | 20.15 | |
| MistralBackbone=Mistral-7B, Training=Standard instruction-tuned2026.05 | 17.69 | 4.13 | 10.2 | 2.3 | 4.8 | 28 | 6.5 | 13 | 15.8 | 53.5 | 12.2 | 30.6 | 17.69 |