Safety on Air-Bench 2024
91.7Safety ScoreConfident Decoding
Evaluation Results
| Method | Links | |
|---|---|---|
| Confident DecodingModel=gpt-oss-20b, Decoding Strategy=Confident (Ours)2026.06 | 91.7 | |
| Confident DecodingModel=gpt-oss-120B, Decoding Strategy=Confident (Ours)2026.06 | 90 | |
| Last LayerModel=gpt-oss-120B, Decoding Strategy=Last Layer2026.06 | 88 | |
| Last LayerModel=gpt-oss-20b, Decoding Strategy=Last Layer2026.06 | 86.7 | |
| Confident DecodingModel=Gemma-4-31B, Decoding Strategy=Confident (Ours)2026.06 | 76 | |
| Last LayerModel=Gemma-4-31B, Decoding Strategy=Last Layer2026.06 | 74 | |
| Confident DecodingModel=Qwen3.5-122B-A10B, Decoding Strategy=Confident (Ours)2026.06 | 67 | |
| Last LayerModel=Qwen3.5-122B-A10B, Decoding Strategy=Last Layer2026.06 | 66 | |
| Last LayerModel=Qwen3.5-27B, Decoding Strategy=Last Layer2026.06 | 65.7 | |
| Confident DecodingModel=Qwen3.5-27B, Decoding Strategy=Confident (Ours)2026.06 | 64.7 | |
| Confident DecodingModel=Qwen3.5-35B-A3B, Decoding Strategy=Confident (Ours)2026.06 | 55.4 | |
| Last LayerModel=Qwen3.5-35B-A3B, Decoding Strategy=Last Layer2026.06 | 51.7 |