Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Safety on Air-Bench 2024
Loading...
91.7
Safety Score
Confident Decoding
50.1
60.9
71.7
82.5
Jun 20, 2026
Safety Score
Updated 1mo ago
Evaluation Results
Method
Method
Links
Safety Score
Confident Decoding
Model=gpt-oss-20b, Dec...
2026.06
91.7
Confident Decoding
Model=gpt-oss-120B, De...
2026.06
90
Last Layer
Model=gpt-oss-120B, De...
2026.06
88
Last Layer
Model=gpt-oss-20b, Dec...
2026.06
86.7
Confident Decoding
Model=Gemma-4-31B, Dec...
2026.06
76
Last Layer
Model=Gemma-4-31B, Dec...
2026.06
74
Confident Decoding
Model=Qwen3.5-122B-A10...
2026.06
67
Last Layer
Model=Qwen3.5-122B-A10...
2026.06
66
Last Layer
Model=Qwen3.5-27B, Dec...
2026.06
65.7
Confident Decoding
Model=Qwen3.5-27B, Dec...
2026.06
64.7
Confident Decoding
Model=Qwen3.5-35B-A3B,...
2026.06
55.4
Last Layer
Model=Qwen3.5-35B-A3B,...
2026.06
51.7
Feedback
Search any
task
Search any
task