Jailbreak Success Rate on flames
38.6Success Rate (Raw)DeepSeek-R1
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| DeepSeek-R1Model=DeepSeek-R12025.11 | 38.6 | 20.5 | 13.4 | 9.2 | |
| GPT-4oModel=GPT-4o2025.11 | 38.5 | 15 | 9.8 | 7 | |
| Qwen3-235BModel=Qwen3-235B2025.11 | 26.6 | 14.6 | 11.2 | 8.2 |
| Method | Links | ||||
|---|---|---|---|---|---|
| DeepSeek-R1Model=DeepSeek-R12025.11 | 38.6 | 20.5 | 13.4 | 9.2 | |
| GPT-4oModel=GPT-4o2025.11 | 38.5 | 15 | 9.8 | 7 | |
| Qwen3-235BModel=Qwen3-235B2025.11 | 26.6 | 14.6 | 11.2 | 8.2 |