Safety Risk Assessment on Held-out Set
10.9Potential RiskInternVL3.5-1B
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| InternVL3.5-1BCategory=Open-source2026.05 | 10.9 | 10.2 | 16.1 | |
| Gemma-3-4BCategory=Open-source2026.05 | 47.9 | 33.3 | 21.2 | |
| Qwen-3-VL-4BCategory=Open-source2026.05 | 58.3 | 53.5 | 48.6 | |
| InternVL3.5-2BCategory=Open-source2026.05 | 59.3 | 18.7 | 16.6 | |
| Qwen-3-VL-2BCategory=Open-source2026.05 | 59.4 | 32.5 | 27.4 | |
| Gemini-2.5-ProCategory=Closed-source2026.05 | 61.4 | 68.3 | 63.8 | |
| Gemma-3-27BCategory=Open-source2026.05 | 65.3 | 53.5 | 26.6 | |
| Gemma-3-12BCategory=Open-source2026.05 | 66.1 | 49.4 | 26.7 | |
| GPT-4o-miniCategory=Closed-source2026.05 | 67.5 | 44.6 | 38.5 | |
| Qwen-3-VL-8BCategory=Open-source2026.05 | 67.7 | 53.3 | 32.4 | |
| Qwen-3-VL-30B-a3bCategory=Open-source2026.05 | 68.1 | 56.3 | 47.9 | |
| EMBGUARD-2BCategory=EMBGUARD2026.05 | 68.3 | 59.5 | 36.6 | |
| GPT-4oCategory=Closed-source2026.05 | 68.6 | 54.5 | 48.3 | |
| GPT-5.1Category=Closed-source2026.05 | 69.1 | 62 | 57 | |
| Gemini-2.5-FlashCategory=Closed-source2026.05 | 70.4 | 68.2 | 64 | |
| EMBGUARD-4BCategory=EMBGUARD2026.05 | 71.2 | 67.6 | 50.1 | |
| Qwen-3-VL-235B-a22bCategory=Open-source2026.05 | 71.3 | 60 | 51.2 | |
| Qwen-3-VL-32BCategory=Open-source2026.05 | 71.8 | 57.2 | 48.8 |