Safety Evaluation on MultiJail English
100Safe Response RateAIM
Evaluation Results
| Method | Links | |
|---|---|---|
| AIMModel=GPT-OSS-120b2025.09 | 100 | |
| Direct InstructionModel=GPT-OSS-120b2025.09 | 99.7 | |
| SteerMoEModel=GPT-OSS-120b2025.09 | 94.3 | |
| SteerMoE + AIMModel=GPT-OSS-120b2025.09 | 9.5 |