MLLM Safety Evaluation on OOD-MMSafe Malicious Mode 1.0 (test)
1.97RA Scoreo4-Mini
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| o4-MiniModel Type=Closed-source2026.03 | 1.97 | 0.2 | 1.98 | 0.9 | 1.99 | 0 | |
| GPT-5.1Model Type=Closed-source2026.03 | 1.96 | 1.5 | 1.96 | 1.3 | 2 | 0 | |
| Qwen3VL-8BModel Type=Open-source2026.03 | 1.91 | 3.7 | 1.92 | 3.3 | 1.98 | 0.4 | |
| Gemini-3-ProModel Type=Closed-source2026.03 | 1.88 | 3.5 | 1.93 | 3.1 | 1.73 | 2.9 | |
| Gemma-3-12BModel Type=Open-source2026.03 | 1.87 | 3.7 | 1.78 | 9.2 | 1.95 | 0.4 | |
| Gemini-3-Pro-ThinkModel Type=Closed-source2026.03 | 1.85 | 5.7 | 1.87 | 5.4 | 1.83 | 1.8 | |
| Qwen3VL-32BModel Type=Open-source2026.03 | 1.85 | 4.8 | 1.9 | 4 | 1.9 | 3.5 | |
| Qwen3VL-4BModel Type=Open-source2026.03 | 1.82 | 8.1 | 1.8 | 8.6 | 1.91 | 3.1 | |
| Doubao-1.6Model Type=Closed-source2026.03 | 1.8 | 6.8 | 1.82 | 7.3 | 1.9 | 1.1 | |
| QwenVL2.5-7BModel Type=Open-source2026.03 | 1.2 | 31 | 1.2 | 37.8 | 1.61 | 7.7 | |
| InternVL-3.5-38BModel Type=Open-source2026.03 | 1.09 | 29 | 1.62 | 17.8 | 1.24 | 18.5 | |
| InternVL-3.5-8BModel Type=Open-source2026.03 | 0.99 | 36.3 | 1.31 | 33 | 1.31 | 12.3 | |
| LLaVA-1.5-7BModel Type=Open-source2026.03 | 0.64 | 54.5 | 0.57 | 70.1 | 1.02 | 22.9 |