Multimodal Question Answering on MME-RealWorld-Lite 1.0 (test)
57.7Perception (AD) AccHART-7B
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| HART-7BParameters=7B, Model Category=Visual Grounded Reasoning Models2026.02 | 57.7 | 49.8 | 89.6 | 58.7 | 86 | 51 | 75 | 72 | 58.7 | 62.4 | |
| Qwen2.5-VL-7B (MGPO + SFT)Parameters=7B, Post-Training Strategy=MGPO + SFT2026.02 | 55.4 | 49.8 | 83.6 | 59.3 | 82 | 47.8 | 71 | 71 | 63.3 | 60.5 | |
| Qwen2.5-VL-7B (GRPO + SFT)Parameters=7B, Post-Training Strategy=GRPO + SFT2026.02 | 49.7 | 47 | 89.2 | 55.3 | 87 | 39 | 74 | 77 | 60 | 58.1 | |
| Qwen2.5-VL-7B (MGPO)Parameters=7B, Post-Training Strategy=MGPO2026.02 | 44 | 46.7 | 86.4 | 54 | 78 | 39.3 | 69 | 74 | 52.7 | 55.1 | |
| Qwen2.5-VL-7B (SFT)Parameters=7B, Post-Training Strategy=SFT2026.02 | 43.7 | 43.6 | 85.6 | 55.3 | 78 | 41 | 60 | 70 | 50.7 | 54 | |
| Qwen2.5-VL-7B (GRPO)Parameters=7B, Post-Training Strategy=GRPO2026.02 | 43.7 | 42.9 | 81.6 | 51.3 | 75 | 38.3 | 53 | 67 | 43.3 | 51.3 | |
| Claude 3.5 SonnetModel Category=Private Models2026.02 | 40.8 | 32.2 | 72.5 | 25.7 | 67.4 | 31.9 | 61.2 | 61.9 | 41.8 | 51.6 | |
| Qwen2.5-VL-32BParameters=32B, Model Category=Larger-scale Open-source Models2026.02 | 40.7 | 29.5 | 87.2 | 40.7 | 83 | 29.5 | 60 | 74 | 27.3 | 45.6 | |
| InternVL3-38BParameters=38B, Model Category=Larger-scale Open-source Models2026.02 | 40 | 42.6 | 85.6 | 56 | 71 | 35 | 45 | 77 | 47.3 | 51 | |
| LLaVA-OneVision-72BParameters=72B, Model Category=Larger-scale Open-source Models2026.02 | 40 | 37.9 | 79.2 | 50.7 | 67 | 39.3 | 41 | 76 | 38.7 | 48.7 | |
| LLaVA-OneVision-7BParameters=7B, Model Category=Open-source General Models2026.02 | 39.4 | 31.7 | 80 | 40 | 56 | 32 | 33 | 65 | 38 | 43.7 | |
| InternVL3-8BParameters=8B, Model Category=Open-source General Models2026.02 | 36.9 | 34.5 | 83.6 | 49.3 | 75 | 37 | 44 | 70 | 40 | 47.9 | |
| DeepEyes-7BParameters=7B, Model Category=Visual Grounded Reasoning Models2026.02 | 33.4 | 43.3 | 90 | 52.7 | 89 | 35 | 69 | 76 | 44 | 53.2 | |
| Pixel-Reasoner-7BParameters=7B, Model Category=Visual Grounded Reasoning Models2026.02 | 30.9 | 38.9 | 89.6 | 52 | 86 | 32.5 | 72 | 71 | 46 | 49.7 | |
| Qwen2.5-VL-72BParameters=72B, Model Category=Larger-scale Open-source Models2026.02 | 30.6 | 27.9 | 90.8 | 34 | 87 | 25.5 | 61 | 74 | 26.7 | 43.7 | |
| Qwen2.5-VL-7BParameters=7B, Model Category=Open-source General Models2026.02 | 30 | 27.3 | 87.6 | 32.7 | 83 | 23 | 62 | 72 | 28.7 | 42.3 | |
| Gemini-1.5-ProModel Category=Private Models2026.02 | 26.6 | 31.1 | 67.6 | 14 | 39.9 | 19.2 | 33.2 | 52.7 | 28.3 | 38.2 | |
| GPT-4o-miniModel Category=Private Models2026.02 | 24.2 | 26.5 | 62.5 | 6.7 | 44.2 | 26.8 | 39.1 | 47 | 25.8 | 36.4 | |
| GPT-4oModel Category=Private Models2026.02 | 22.4 | 33.9 | 77.7 | 28.9 | 46.7 | 26.4 | 44.8 | 61.4 | 36.5 | 45.2 |