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

Jailbreak Robustness on HarmBench (Cost & Efficiency Metrics)

0.5AE (Risk/TFLOPs) GCG

Tulu3 SFT

-0.5046.27313.0519.827Jun 9, 2026
Updated 1mo ago

Evaluation Results

MethodLinks
2026.06
0.5--52.43.58.9314250
2026.06
0.9-31.321.216.622.1369886
2026.06
1521.279.940.9610.4527567
2026.06
1503.672.425.76.718.9547990
2026.06
1.3399.738.922.813.623739794
2026.06
2.118944.824.57.616677583
2026.06
3.3173.733.913.415.936.8819798
2026.06
4.9-----15--
2026.06
8.459.311.29.23953.3100100100
2026.06
25.62015.58.230.659.6999999