Safety Evaluation on Harm Categories evaluated by Claude-Opus-4-7 (Average)
95.3Safety ScoreSafeMoE-XL
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SafeMoE-XLBackbone=Mistral-7B, LLM Judge=Claude-Opus-4-7, Seeds=32026.05 | 95.3 | 8.13 | |
| SafeMoE-LBackbone=Mistral-7B, LLM Judge=Claude-Opus-4-7, Seeds=32026.05 | 91.4 | 7.08 | |
| SafeMoE-4Backbone=Mistral-7B, LLM Judge=Claude-Opus-4-7, Seeds=32026.05 | 89.6 | 6.61 | |
| SafeMoE-QwenBackbone=Qwen-3B, LLM Judge=Claude-Opus-4-7, Seeds=32026.05 | 67.2 | 6.03 | |
| Mistral-7BBackbone=Mistral-7B, LLM Judge=Claude-Opus-4-7, Seeds=32026.05 | 36.4 | 2.73 | |
| QwenBackbone=Qwen-3B, LLM Judge=Claude-Opus-4-7, Seeds=32026.05 | 21 | 4.43 |