Jailbreak Attack on HarmBench Standard set
98CodeChaDeepSeek-v3
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| DeepSeek-v3Family=DeepSeek, Type=Instruction2026.03 | 98 | 91 | 61 | 70.5 | 68.5 | 99 | 80 | 99.5 | |
| GPT-4.1-miniFamily=OpenAI, Type=Instruction2026.03 | 95.5 | 83.5 | 47.5 | 58 | 12 | 96 | 70 | 99.5 | |
| GPT-4oFamily=OpenAI, Type=Instruction2026.03 | 91 | 77.5 | 36 | 57 | 33.5 | 96.5 | 54 | 95 | |
| Llama3.3-70B-InstructFamily=Llama, Type=Instruction2026.03 | 90.5 | 62 | 59 | 68 | 0 | 89 | 75 | 85 | |
| Qwen2.5-72B-InstructFamily=Qwen, Type=Instruction2026.03 | 90.5 | 81.5 | 51 | 61 | 2 | 98 | 80 | 98.5 | |
| Claude-3-7-SonnetFamily=Claude, Type=Instruction2026.03 | 86.5 | 66 | 12.5 | 51 | 3 | 2 | 17 | 85.5 | |
| Phi-4Family=Phi, Type=Instruction2026.03 | 60 | 60.5 | 49 | 33 | 6 | 62.5 | 38.5 | 93.5 | |
| GPT-OSS-20BFamily=OpenAI, Type=Reasoning2026.03 | 41.5 | 40 | 30 | 27 | 1.5 | 82.5 | 6.5 | 96 |