Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

Jailbreaking on HarmBench Open-Source Models

81.5Llama-3.1-8B Success Rate

DIALTREE

8.727.646.565.4Oct 2, 2025
Updated 1mo ago

Evaluation Results

MethodLinks
2025.10
81.589.58588.58353.580.2
2025.10
43506748342945.2
2025.10
4148.5534043.51.537.9
2025.10
38.543.534.53637.51033.3
2025.10
33.525.541.515.515322.3
2025.10
29.533.548.52420.53.526.6
2025.10
172234.5125.5215.5
2025.10
1228.518.514.5183220.6
2025.10
11.58.54321.519.5017.3