AdvBench
Benchmarks
Task NameDataset NameSOTA ResultTrendResults
AdvBench Anchoring (test)
92ASR (GPT-4o)
5
AdvBench
0Harmful Rate
4
AdvBench
1.9Time/Sample (min)
4
AdvBench
97.5Refusal Rate
4
AdvBench
29Llama-2 ASR
4
AdvBench
21ASR (Llama-2)
4
AdvBench All targets (150 goals)
98ASR
4
AdvBench minimax-M2.5 (50 goals)
100Attack Success Rate (ASR)
4
AdvBench Base64 prompts llama-3.1-8b
67.9Attack Success Rate (ASR)
4
AdvBench Zulu prompts llama-3.1-8b
66.8Attack Success Rate (ASR)
4
AdvBench Puzzler prompts llama-3.1-8b
100ASR
4
AdvBench DrAttack prompts llama-3.1-8b
74.1Attack Success Rate (ASR)
4
AdvBench IJP prompts llama-3.1-8b
48.9ASR
4
AdvBench
98StrongREJECT Score
4
AdvBench GPT-4O-MINI
55.6ASR@1
4
AdvBench GPT-3.5-TURBO
61.4ASR@1
4
AdvBench (test)
91.3ASR
4
AdvBench
91.8ASR
4
AdvBench-50
0Unsafe Interaction Rate
4
AdvBench unsafe inputs
94ASR
4
AdvBench
0ASR
4
AdvBench
32Attack Success Rate (ASR)
3
AdvBench 2023 (test)
64.9Attack Success Rate (ASP)
3
AdvBench
41PAR
3
AdvBench
2.69ASR (Overall)
3