Motion Understanding on MotionBench
65.4AccuracyGPT-5.2
Evaluation results
| Method | Links | |
|---|---|---|
| GPT-5.2Model Class=Closed-source Models2026.06 | 65.4 | |
| GOPAgenModel Category=Agentic video understanding framework2026.06 | 65.3 | |
| Deep Video DiscoveryModel Category=Agentic video understanding framework2026.06 | 63.4 | |
| Gemini 3 ProModel Class=Closed-source Models2026.06 | 62.6 | |
| Qwen3-VL-8B + MotionAtlas-DataModel Class=Open-source Models, Training Data=MotionAtlas-Data2026.06 | 62.6 | |
| InternVL3.5Size=8B, mode=instruct2026.04 | 62.3 | |
| Gemini 2.5 ProModel Class=Closed-source Models2026.06 | 62 | |
| Qwen3-VL-4B + MotionAtlas-DataModel Class=Open-source Models, Training Data=MotionAtlas-Data2026.06 | 61.9 | |
| Molmo2-8B + MotionAtlas-DataModel Class=Open-source Models, Training Data=MotionAtlas-Data2026.06 | 61.8 | |
| Qwen3-OmniSize=30B-A3B, mode=instruct2026.04 | 61.7 | |
| Qwen-2.5-VL-72BMode=Zero-shot2025.12 | 61.4 | |
| MiniCPM-o 4.5Size=9B, mode=instruct2026.04 | 61.4 | |
| InternVL2.5-72BModel Category=Open-Source Models2026.06 | 61 | |
| Molmo2-4B + MotionAtlas-DataModel Class=Open-source Models, Training Data=MotionAtlas-Data2026.06 | 60.6 | |
| Molmo2-8BModel Class=Open-source Models2026.06 | 60.5 | |
| Molmo2-4BModel Class=Open-source Models2026.06 | 59.9 | |
| Qwen3-VLSize=8B, mode=instruct2026.04 | 59.5 | |
| VideoLucyModel Category=Agentic video understanding framework2026.06 | 59.1 | |
| Qwen3-VL-8BModel Class=Open-source Models2026.06 | 59 | |
| Qwen2.5-VL-72BModel Category=Open-Source Models2026.06 | 56.1 | |
| Qwen3-VL-4BModel Class=Open-source Models2026.06 | 55.9 | |
| Gemini-2.5-FlashMode=Zero-shot2025.12 | 55.6 | |
| VideoTreeModel Category=Agentic video understanding framework2026.06 | 53.1 | |
| VCAModel Category=Agentic video understanding framework2026.06 | 51.3 | |
| Gemini-1.5-ProModel Category=Closed-Source Models2026.06 | 50 | |
| DeepSportFrames=8.32025.11 | 48.5 | |
| NVILA-Video-15BMode=Fine-tuned, Data=PLM 467k2025.12 | 47.5 | |
| NVILA-Video-15BMode=Fine-tuned, Data=FoundationMotion2025.12 | 46.7 | |
| NVILA-Video-15BMode=Zero-shot2025.12 | 45.7 | |
| NVILA-Video-8BMode=Fine-tuned, Data=PLM 467k2025.12 | 43.6 | |
| Qwen2.5-VL-7B-InstructFrames=16.02025.11 | 43.1 | |
| NVILA-Video-8BMode=Fine-tuned, Data=FoundationMotion2025.12 | 42.9 | |
| NVILA-Video-8BMode=Zero-shot2025.12 | 42.3 | |
| Qwen-2.5-VL-7BMode=Fine-tuned, Data=FoundationMotion2025.12 | 41.3 | |
| Qwen-2.5-VL-7BMode=Zero-shot2025.12 | 39.1 |