Jailbreaking on HarmBench Textual Replacement
55.4ASR (K=1)Gemini 3 Flash
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Gemini 3 Flash2026.05 | 55.4 | 58.8 | |
| Qwen3-VL-32B2026.05 | 35.6 | 39 | |
| Qwen3-VL-235B2026.05 | 25.3 | 29.5 | |
| Gemini 3.1 Pro2026.05 | 18.4 | 19 | |
| GPT-5.22026.05 | 12.8 | 16.9 | |
| Claude Haiku 4.52026.05 | 6.1 | 8.1 |