Video Motion Reasoning on MotionBench (dev)
63Overall AccuracyOpen-o3 + MCoT
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Open-o3 + MCoTModel Category=Ours, Configuration=w/ r_ground2026.03 | 63 | — | — | — | — | — | — | |
| Motion-oModel Category=Ours, Configuration=full2026.03 | 60 | — | — | — | — | — | — | |
| Open-o3 + MCoTModel Category=Ours2026.03 | 60 | — | — | — | — | — | — | |
| Motion-oModel Category=Ours, Configuration=w/o r_ground2026.03 | 55 | — | — | — | — | — | — | |
| VideoLLaMA3-2B + FAVOR-Train + MOSSTraining=Fine-tuned on FAVOR-Train, Module=MOSS2026.04 | 54.2 | 59.7 | 55.7 | 48.3 | 68.1 | 38.5 | 34 | |
| VideoLLaMA3-2B + FAVOR-TrainTraining=Fine-tuned on FAVOR-Train2026.04 | 51.4 | 55.8 | 54.4 | 36.4 | 68.8 | 38.3 | 33 | |
| VideoLLaMA3-2BTraining=Zero-shot / Pre-trained2026.04 | 50.2 | 54.9 | 54.2 | 36.1 | 68.3 | 37 | 27.3 | |
| Open-o3 Video-7BModel Category=Open-source2026.03 | 45 | — | — | — | — | — | — | |
| Qwen2.5-VL-7BModel Category=Open-source2026.03 | 35 | — | — | — | — | — | — | |
| GPT-4oModel Category=Proprietary2026.03 | 33 | — | — | — | — | — | — |