Binary Harmfulness Detection on BLM-Guard
93.1NR (B)InternVL3
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| InternVL3Parameters=14B2026.02 | 93.1 | 51.9 | 48.5 | 95 | 64.2 | |
| InternVL3Parameters=8B2026.02 | 92.1 | 29.9 | 46.1 | 96.1 | 62.3 | |
| Kimi-VL-A3B-ThinkingParameters=A3B, Variant=Thinking2026.02 | 88.7 | 56.1 | 58.8 | 89.8 | 71.1 | |
| Qwen2.5-VLParameters=32B2026.02 | 87.5 | 82.3 | 76.9 | 83.5 | 80.1 | |
| QwenGuardParameters=7B2026.02 | 82.3 | 69.5 | 64.2 | 78.5 | 70.6 | |
| BLM-Guard2026.02 | 78.1 | 95 | 97.6 | 96.2 | 96.9 | |
| Qwen2.5-VLParameters=7B2026.02 | 75.5 | 91.8 | 83.1 | 57.6 | 68 | |
| LLaVA-Next-VideoParameters=7B2026.02 | 34.8 | 31.01 | 15.4 | 17.8 | 16.5 | |
| LlavaGuardParameters=7B2026.02 | 32.9 | 25.8 | 19.3 | 25.1 | 21.8 |