Attack Success Rate on 20,000 harmful requests and 20,000 jailbreak prompts (test)
80.2Attack Success Rate (ASR)GateBreaker
Evaluation Results
| Method | Links | |
|---|---|---|
| GateBreakerTarget Model=GPT-OSS-20B2025.12 | 80.2 | |
| GateBreakerTarget Model=Hunyuan-A13B-Instruct2025.12 | 76.9 | |
| GateBreakerTarget Model=OpenPangu-Pro-MoE-72B2025.12 | 73.1 | |
| GateBreakerTarget Model=Average2025.12 | 64.9 | |
| GateBreakerTarget Model=Mixtral-8x7B-Instruct-v0.12025.12 | 64.2 | |
| GateBreakerTarget Model=Qwen1.5-MoE-A2.7B-Chat2025.12 | 57.8 | |
| GateBreakerTarget Model=Qwen3-30B-A3B-Instruct-25072025.12 | 56.9 | |
| GateBreakerTarget Model=Phi-3.5-MoE-Instruct2025.12 | 56.5 | |
| GateBreakerTarget Model=Deepseek-MoE-16b-Chat2025.12 | 53.6 | |
| SAFExTarget Model=Mixtral-8x7B-Instruct-v0.12025.12 | 48.8 | |
| SAFExTarget Model=Deepseek-MoE-16b-Chat2025.12 | 35.6 | |
| SAFExTarget Model=Qwen1.5-MoE-A2.7B-Chat2025.12 | 35 | |
| SAFExTarget Model=OpenPangu-Pro-MoE-72B2025.12 | 30 | |
| SAFExTarget Model=Average2025.12 | 29.9 | |
| SAFExTarget Model=Qwen3-30B-A3B-Instruct-25072025.12 | 28.4 | |
| SAFExTarget Model=Hunyuan-A13B-Instruct2025.12 | 28.4 | |
| SAFExTarget Model=Phi-3.5-MoE-Instruct2025.12 | 26.8 | |
| SAFExTarget Model=GPT-OSS-20B2025.12 | 6.4 |