Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Toxicity Detection on Safety Evaluation Set Toxicity (held-out target labels)
Loading...
97.6
AUROC
Safety Monitor (In-model)
56.416
67.108
77.8
88.492
May 12, 2026
AUROC
Updated 21d ago
Evaluation Results
Method
Method
Links
AUROC
Safety Monitor (In-model)
Model=Qwen-2.5-7B-Inst...
2026.05
97.6
Quotient Transfer
Target Model=Qwen-2.5-...
2026.05
97.1
Quotient Transfer
Target Model=Qwen-14B,...
2026.05
95.5
Quotient Transfer
Target Model=Qwen-3B,...
2026.05
95.2
Quotient Transfer
Target Model=Mistral,...
2026.05
92.6
Random-initialization control
2026.05
58
Feedback
Search any
task
Search any
task