Jailbreaking on HarmBench HADES
1ASR (K=1)GPT-5.2
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GPT-5.22026.05 | 1 | 2 | |
| Claude Haiku 4.52026.05 | 6 | 9 | |
| Qwen3-VL-235B2026.05 | 8 | 11 | |
| Gemini 3 Flash2026.05 | 9 | 12 | |
| Gemini 3.1 Pro2026.05 | 11 | 13.1 | |
| Qwen3-VL-32B2026.05 | 25 | 32 |
| Method | Links | ||
|---|---|---|---|
| GPT-5.22026.05 | 1 | 2 | |
| Claude Haiku 4.52026.05 | 6 | 9 | |
| Qwen3-VL-235B2026.05 | 8 | 11 | |
| Gemini 3 Flash2026.05 | 9 | 12 | |
| Gemini 3.1 Pro2026.05 | 11 | 13.1 | |
| Qwen3-VL-32B2026.05 | 25 | 32 |