Jailbreak Attack Efficiency on AdvBench (100-instance sample)
2.6Average QueriesSalami
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SalamiTarget Model=GLM-4.52026.04 | 2.6 | 508.44 | 53.72 | |
| SalamiTarget Model=GPT-4o2026.04 | 3.9 | 728.91 | 88.35 | |
| CrescendoTarget Model=GLM-4.52026.04 | 4.3 | 3,182.78 | 109.68 | |
| CrescendoTarget Model=GPT-4o2026.04 | 5.6 | 5,013.49 | 168.83 | |
| PAIRTarget Model=GLM-4.52026.04 | 9.2 | 5,059.1 | 187.1 | |
| PAIRTarget Model=GPT-4o2026.04 | 10.7 | 6,221.29 | 202.75 |