Safety Detection on MAS Qwen-2.5-7B (Current)
97.46AccuracyAgentLens
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| AgentLensMethod Category=Interpretability-based2026.06 | 97.46 | 96.25 | |
| LLaMA-Guard-4-12BMethod Category=Guardrail-based2026.06 | 94.92 | 92.77 | |
| Deepseek-V4-ProMethod Category=LLM-as-a-Judge2026.06 | 90.68 | 88.17 | |
| Gemini-3-ProMethod Category=LLM-as-a-Judge2026.06 | 90.25 | 87.7 | |
| Deepseek-V4-FlashMethod Category=LLM-as-a-Judge2026.06 | 89.83 | 87.1 | |
| Claude Sonnet-4.6Method Category=LLM-as-a-Judge2026.06 | 87.29 | 84.54 | |
| AGrail (GPT-4o)Method Category=Guardrail-based2026.06 | 85.17 | 78.76 | |
| Claude Opus-4.6Method Category=LLM-as-a-Judge2026.06 | 82.63 | 80 | |
| GPT-4oMethod Category=LLM-as-a-Judge2026.06 | 68.22 | 68.62 | |
| GPT-5.2Method Category=LLM-as-a-Judge2026.06 | 49.15 | 57.75 |