Safety Evaluation on Jailbreaking Scenarios (Jailbreak)
2.3Unsafe RatioDeepSeek-V3
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DeepSeek-V3risk control system=true2025.01 | 2.3 | 46.5 | — | |
| DeepSeek-R1risk control system=true2025.01 | 4.3 | 87.3 | — | |
| o1release date=2024-12-172025.01 | 12.1 | 79.8 | — | |
| Claude-3.7-Sonnet2025.01 | 26.2 | 21.9 | — | |
| Qwen2.5 Instructparameters=72B2025.01 | 29.7 | 25.2 | — | |
| GPT-4orelease date=2024-05-132025.01 | 30.4 | 57.3 | — | |
| DeepSeek-V3risk control system=false2025.01 | 36.4 | 8.9 | — | |
| DeepSeek-R1risk control system=false2025.01 | 85.9 | 1.9 | — |