Jailbreak Attack on MM-SafetyBench (tiny)
99.16ASRMIDAS
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MIDASTarget Model=Gemini-2.5-FT2026.02 | 99.16 | 4.35 | |
| MIDASTarget Model=QVQ-Max2026.02 | 98.65 | 4.21 | |
| MIDASTarget Model=Gemini-2.5-Pro2026.02 | 92.17 | 3.94 | |
| VisCRATarget Model=QVQ-Max2026.02 | 82.2 | 3.53 | |
| MIDASTarget Model=GPT-5-Chat2026.02 | 81.54 | 3.49 | |
| MIDASTarget Model=GPT-4o2026.02 | 61.07 | 2.53 | |
| VisCRATarget Model=Gemini-2.5-FT2026.02 | 49.7 | 2.29 | |
| HIMRDTarget Model=Gemini-2.5-Pro2026.02 | 38.3 | 1.38 | |
| VisCRATarget Model=GPT-4o2026.02 | 37.12 | 1.68 | |
| VisCRATarget Model=Gemini-2.5-Pro2026.02 | 35.92 | 1.69 | |
| HIMRDTarget Model=GPT-4o2026.02 | 26.4 | 0.95 | |
| HIMRDTarget Model=GPT-5-Chat2026.02 | 26.4 | 0.97 | |
| FigStepTarget Model=QVQ-Max2026.02 | 22.78 | 0.94 | |
| FigStepTarget Model=Gemini-2.5-FT2026.02 | 20.56 | 0.71 | |
| VisCRATarget Model=GPT-5-Chat2026.02 | 20.24 | 0.97 | |
| HIMRDTarget Model=QVQ-Max2026.02 | 16.8 | 1.1 | |
| SI-AttackTarget Model=QVQ-Max2026.02 | 13.94 | 1.29 | |
| FigStepTarget Model=GPT-4o2026.02 | 11.93 | 0.51 | |
| FigStepTarget Model=Gemini-2.5-Pro2026.02 | 11.82 | 0.46 | |
| FigStepTarget Model=GPT-5-Chat2026.02 | 11.82 | 0.49 | |
| HIMRDTarget Model=Gemini-2.5-FT2026.02 | 10.2 | 0.67 | |
| SI-AttackTarget Model=Gemini-2.5-FT2026.02 | 5.52 | 0.89 | |
| SI-AttackTarget Model=GPT-5-Chat2026.02 | 5.52 | 0.63 | |
| SI-AttackTarget Model=Gemini-2.5-Pro2026.02 | 0.6 | 0.86 | |
| SI-AttackTarget Model=GPT-4o2026.02 | 0 | 0.38 |