Trajectory-level safety evaluation on ASSE-Safety (test)
81.1AccuracyAgentDoG-Llama3.1-8B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| AgentDoG-Llama3.1-8BModel Category=Our Models2026.01 | 81.1 | 80 | 87.2 | 83.4 | |
| AgentDoG-Qwen3-4BModel Category=Our Models2026.01 | 80.4 | 85.5 | 77.3 | 81.2 | |
| AgentDoG-Qwen2.5-7BModel Category=Our Models2026.01 | 79.8 | 84.3 | 77.4 | 80.7 | |
| ShieldAgentModel Category=Guard Models2026.01 | 79.6 | 76.5 | 90.5 | 82.9 | |
| Gemini-3-ProModel Category=Closed-Source2026.01 | 78.5 | 85.9 | 72.5 | 78.6 | |
| Qwen3-235B-A22B-Instruct-2507Model Category=Open-Source2026.01 | 77.6 | 85.4 | 71.1 | 77.6 | |
| GPT-5.2Model Category=Closed-Source2026.01 | 77.4 | 82.4 | 74.4 | 78.2 | |
| Gemini-3-FlashModel Category=Closed-Source2026.01 | 75.9 | 89.5 | 63.4 | 74.2 | |
| QwQ-32BModel Category=Open-Source2026.01 | 68.2 | 88.7 | 47.8 | 62.1 | |
| Qwen3-4B-Instruct-2507Model Category=Open-Source2026.01 | 60.9 | 87.5 | 33.1 | 48.1 | |
| Qwen2.5-7B-InstructModel Category=Open-Source2026.01 | 58.8 | 81.3 | 31.9 | 45.8 | |
| LlamaGuard4-12BModel Category=Guard Models2026.01 | 56.3 | 79.9 | 27.8 | 41.2 | |
| PolyGuardModel Category=Guard Models2026.01 | 56.3 | 62.7 | 49.4 | 55.2 | |
| Llama3.1-8B-InstructModel Category=Open-Source2026.01 | 55.2 | 55 | 98.3 | 70.6 | |
| LlamaGuard3-8BModel Category=Guard Models2026.01 | 54.5 | 77.6 | 24.7 | 37.4 | |
| Qwen3-GuardModel Category=Guard Models2026.01 | 48.2 | 61.2 | 15.8 | 25.1 | |
| ShieldGemma-27BModel Category=Guard Models2026.01 | 47.2 | 65.1 | 7 | 12.6 | |
| ShieldGemma-9BModel Category=Guard Models2026.01 | 47.2 | 71 | 5.5 | 10.1 | |
| NemoGuardModel Category=Guard Models2026.01 | 43.4 | 47.2 | 30.3 | 36.9 | |
| JoySafetyModel Category=Guard Models2026.01 | 39.9 | 40.9 | 22.6 | 29.1 |