Severity-wise Harmfulness Classification on BLM-Guard
97.8Accuracy (High)BLM-Guard
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| BLM-Guard2026.02 | 97.8 | 94 | 90.2 | 98.6 | 96.2 | 91.4 | 84.5 | |
| Qwen2.5-VLParameters=7B2026.02 | 74.5 | 70.1 | 66.9 | 78.1 | 71.2 | 70.1 | 64.2 | |
| QwenGuardParameters=7B2026.02 | 73.3 | 63.1 | 59.9 | 69.3 | 58.2 | 51.2 | 59.8 | |
| Qwen2.5-VLParameters=32B2026.02 | 73.2 | 68.5 | 65.2 | 76.8 | 70.3 | 68.2 | 66.7 | |
| Kimi-VL-A3B-ThinkingParameters=A3B, Variant=Thinking2026.02 | 72.8 | 68.4 | 65.1 | 62.4 | 52.9 | 51.1 | 70.1 | |
| InternVL3Parameters=8B2026.02 | 69.5 | 64.5 | 60.7 | 70.1 | 38.9 | 63.1 | 61.1 | |
| InternVL3Parameters=14B2026.02 | 58.1 | 53.5 | 50.2 | 62.2 | 52.1 | 49.9 | 59.3 | |
| LLaVA-Next-VideoParameters=7B2026.02 | 32.1 | 34.5 | 43.5 | 44.3 | 34.2 | 31.2 | 53.3 | |
| LlavaGuardParameters=7B2026.02 | 30.1 | 32.2 | 39.5 | 39.2 | 25.5 | 23.5 | 52.7 |