HarmBench
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
HarmBench
100Attack Success Rate (ASR)
624
HarmBench (test)
100ASRHB
276
HarmBench
96.3ASR
244
HarmBench
0ASR
153
Harmbench
0.06Harmbench Score
127
HarmBench
84Attack Success Rate (ASR)
110
HarmBench
98.94F1 Score
100
HarmBench
0PAIR ASR
91
HarmBench
0ASR
88
HarmBench
1Unsafe Rate
72
HarmBench
0HarmBench ASR
72
HarmBench
0ASR
62
HarmBench
0Harmful Score (%)
60
HarmBench
97Attack Success Rate (Generated)
55
HARMBENCH 159 standard behaviors (test)
0ASR
55
HarmBench
0ASR
52
HarmBench
1.01Toxicity Score
50
HarmBench Standard Behaviours (200 examples)
0ASR
48
HarmBench-191 (dev)
97.4Attack Success Rate (ASR)
42
HarmBench single-turn
91ASR
41
HARMBENCH
96.27Attack Success Rate (ASR)
40
HarmBench
87.5HarmBench Score
40
Harmbench
3.5ASR
39
HarmBench
93.25PAIR
39
HarmBench
96Refusal Rate
38