Jailbreaking on HarmBench SD
6.3ASR (K=1)Claude Haiku 4.5
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Claude Haiku 4.52026.05 | 6.3 | 6.3 | |
| GPT-5.22026.05 | 6.3 | 10.8 | |
| Gemini 3.1 Pro2026.05 | 7.6 | 7.6 | |
| Gemini 3 Flash2026.05 | 22.2 | 22.2 | |
| Qwen3-VL-235B2026.05 | 32.9 | 48.7 | |
| Qwen3-VL-32B2026.05 | 41.1 | 56.3 |