List your company

Safety on Air-Bench 2024

91.7Safety ScoreConfident Decoding
50.160.971.782.5Jun 20, 2026

Evaluation results

MethodLinks
Confident DecodingModel=gpt-oss-20b, Decoding Strategy=Confident (Ours)2026.0691.7
Confident DecodingModel=gpt-oss-120B, Decoding Strategy=Confident (Ours)2026.0690
Last LayerModel=gpt-oss-120B, Decoding Strategy=Last Layer2026.0688
Last LayerModel=gpt-oss-20b, Decoding Strategy=Last Layer2026.0686.7
Confident DecodingModel=Gemma-4-31B, Decoding Strategy=Confident (Ours)2026.0676
Last LayerModel=Gemma-4-31B, Decoding Strategy=Last Layer2026.0674
Confident DecodingModel=Qwen3.5-122B-A10B, Decoding Strategy=Confident (Ours)2026.0667
Last LayerModel=Qwen3.5-122B-A10B, Decoding Strategy=Last Layer2026.0666
Last LayerModel=Qwen3.5-27B, Decoding Strategy=Last Layer2026.0665.7
Confident DecodingModel=Qwen3.5-27B, Decoding Strategy=Confident (Ours)2026.0664.7
Confident DecodingModel=Qwen3.5-35B-A3B, Decoding Strategy=Confident (Ours)2026.0655.4
Last LayerModel=Qwen3.5-35B-A3B, Decoding Strategy=Last Layer2026.0651.7