Attack Success Rate on SALAD-Bench Heavy Blur
0Attack Success Rate (ASR)GPT-4o
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GPT-4oOptimization State=Before2026.04 | 0 | — | |
| Claude Sonnet 4.5Optimization State=Before2026.04 | 0 | — | |
| GPT-4oOptimization State=After2026.04 | 10 | — | |
| Claude Sonnet 4.5Optimization State=After2026.04 | 28 | 28 | |
| Qwen3-VL-4BOptimization State=Before2026.04 | 44 | — | |
| Mistral-Large-3Optimization State=Before2026.04 | 52 | — | |
| Qwen3-VL-4BOptimization State=After2026.04 | 54 | — | |
| Mistral-Large-3Optimization State=After2026.04 | 72 | — |