Content Moderation on MMDS (test)
95.76AccuracyLLaVAShield-7B
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| LLaVAShield-7BOpen=true2025.09 | 95.76 | 91.76 | 100 | 95.71 | |
| LLaVAShield-7BOpen=true2025.09 | 94.24 | 87.6 | 97.41 | 92.24 | |
| GPT-5-miniOpen=false2025.09 | 85.76 | 64.34 | 98.81 | 77.93 | |
| Gemini-2.5-ProOpen=false2025.09 | 80 | 48.84 | 100 | 65.62 | |
| GPT-5-miniOpen=false2025.09 | 79.7 | 60.59 | 100 | 75.46 | |
| GPT-4oOpen=false2025.09 | 76.67 | 41.09 | 98.15 | 57.92 | |
| Qwen3-VL-30B-A3B-InstructOpen=true2025.09 | 75.76 | 40.31 | 94.55 | 56.52 | |
| Gemini-2.5-ProOpen=false2025.09 | 72.73 | 47.06 | 100 | 64 | |
| GPT-4oOpen=false2025.09 | 71.21 | 44.71 | 98.7 | 61.54 | |
| InternVL3.5-38BOpen=true2025.09 | 69.7 | 22.48 | 100 | 36.71 | |
| OpenAI ModerationOpen=false2025.09 | 67.58 | 23.26 | 78.95 | 35.93 | |
| Qwen2.5-VL-72B-InstructOpen=true2025.09 | 67.27 | 16.28 | 100 | 28 | |
| Llama Guard-4-12BOpen=true2025.09 | 66.06 | 17.05 | 81.48 | 28.21 | |
| Ovis2-34BOpen=true2025.09 | 63.64 | 7.75 | 90.91 | 16.46 | |
| InternVL3-78BOpen=true2025.09 | 63.03 | 5.43 | 100 | 10.29 | |
| InternVL3-8BOpen=true2025.09 | 62.12 | 3.88 | 83.33 | 7.41 | |
| Qwen2.5-VL-7B-InstructOpen=true2025.09 | 61.21 | 0.78 | 100 | 1.54 | |
| LLaVA-OV-7BOpen=true2025.09 | 60.91 | 0 | 0 | 0 | |
| InternVL3-78BOpen=true2025.09 | 60.03 | 22.94 | 100 | 37.32 | |
| Qwen2.5-VL-72B-InstructOpen=true2025.09 | 58.79 | 20 | 100 | 33.33 | |
| InternVL3.5-38BOpen=true2025.09 | 57.27 | 17.06 | 100 | 29.15 | |
| Qwen3-VL-30B-A3B-InstructOpen=true2025.09 | 54.55 | 11.76 | 100 | 21.05 | |
| Ovis2-34BOpen=true2025.09 | 53.33 | 10 | 94.44 | 18.09 | |
| Llama Guard-4-12BOpen=true2025.09 | 52.42 | 7.65 | 100 | 14.21 | |
| LLaVA-OV-7BOpen=true2025.09 | 48.79 | 0.59 | 100 | 1.17 | |
| Qwen2.5-VL-7B-InstructOpen=true2025.09 | 48.79 | 0.59 | 100 | 1.17 | |
| InternVL3-8BOpen=true2025.09 | 48.48 | 0 | 0 | 0 |