Safety Evaluation on XSTest Unsafe (Safe/Info Metrics)
90Safe RateMistral MoE-XL
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Mistral MoE-XLModel Type=SafeMoE, Base Model=Mistral-7B2026.05 | 90 | 6.2 | |
| Qwen MoE-XLModel Type=SafeMoE, Base Model=Qwen-3B2026.05 | 47.1 | 7.1 | |
| Qwen-3BModel Type=standard instruction-tuned2026.05 | 12.3 | 7.17 | |
| MistralModel Type=standard instruction-tuned2026.05 | 8.5 | 5.6 |