Risk Identification on SafeAgentBench
80.77RIRHomeGuard-8B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| HomeGuard-8BModel Category=Ours, Parameters=8B2026.03 | 80.77 | 30.69 | |
| PoexModel Category=Runtime Safeguard2026.03 | 76.54 | 31.41 | |
| ThinkSafeModel Category=Runtime Safeguard2026.03 | 75.49 | 32.85 | |
| Gemini-3-ProModel Category=Proprietary Models2026.03 | 72.26 | 32.49 | |
| HomeGuard-4BModel Category=Ours, Parameters=4B2026.03 | 71.08 | 20.58 | |
| RoboBrain2.5-8BModel Category=Open-Sourced VLMs2026.03 | 70.37 | 47.65 | |
| Qwen3-VL-235B-A22B-ThinkingModel Category=Open-Sourced VLMs2026.03 | 68.9 | 33.21 | |
| GPT-4o-miniModel Category=Proprietary Models2026.03 | 65.78 | 65.7 | |
| Qwen3-VL-4B-ThinkingModel Category=Open-Sourced VLMs2026.03 | 63.67 | 36.46 | |
| Qwen3-VL-8B-ThinkingModel Category=Open-Sourced VLMs2026.03 | 63.49 | 43.68 | |
| Gemini-2.5-flashModel Category=Proprietary Models2026.03 | 63.32 | 34.3 | |
| AgentSpecModel Category=Runtime Safeguard2026.03 | 61.9 | 42.24 |