Share your thoughts, 1 month free Claude Pro on us
See more
Home
/
Benchmarks
Safety Evaluation on AIR-Bench 64
Loading...
3.12
Harm Score
SFT
3.0464
3.5432
4.04
4.5368
Oct 17, 2025
Harm Score
Utility Score
Accuracy
Updated 1mo ago
Evaluation Results
Method
Method
Links
Harm Score
Utility Score
Accuracy
SFT
Strategy=Threshold (τ...
2025.10
3.12
7.75
72.47
RLVR
Strategy=Threshold (τ...
2025.10
3.33
7.4
79.6
SFT
Strategy=Top-k (k = 100)
2025.10
4.49
7.78
65.73
RLVR
Strategy=Top-k (k = 100)
2025.10
4.68
7.45
79.27
SFT
Strategy=Baseline
2025.10
4.81
7.44
66.13
RLVR
Strategy=Baseline
2025.10
4.96
7.73
80.93
Feedback
Search any
task
Search any
task
