Long Video Understanding on VideoMME Long split (30-60 min)
65.3AccuracyGPT-4o
Evaluation results
| Method | Links | |
|---|---|---|
| GPT-4oInput Frames=>=642025.12 | 65.3 | |
| VideoMem*Size=8B, Input Frames=642025.12 | 64.2 | |
| Qwen2.5-VLSize=72B, Input Frames=>=642025.12 | 61.2 | |
| Qwen3-VL*Size=8B, Input Frames=642025.12 | 58.6 | |
| GPT-4VInput Frames=>=642025.12 | 56.9 | |
| Thinking-QwenVLFrames=1fps (100%)2026.04 | 56.4 | |
| NVILASize=8B, Input Frames=>=642025.12 | 54.8 | |
| Flow4AgentSize=7B, Input Frames=>=642025.12 | 54.2 | |
| VideoLLaMA3Size=7B, Input Frames=>=642025.12 | 54.1 | |
| Qwen2.5-VLSize=7B, Input Frames=>=642025.12 | 51.6 | |
| LLaVA-videoSize=7B, Input Frames=>=642025.12 | 50.6 | |
| Qwen2.5-VL-7BFrames=1fps2026.04 | 50.4 | |
| Video-XL-7BFrames=1282026.04 | 49.2 | |
| VISTA-7BFrames=-2026.04 | 49.2 | |
| Thinking-QwenVLFrames=1fps (↓93.75%)2026.04 | 49.1 | |
| TimeChat-Online-7BFrames=1fps (100%)2026.04 | 48.4 | |
| LongVASize=7B, Input Frames=>=642025.12 | 47.6 | |
| LLaVA-OnevisionSize=7B, Input Frames=>=642025.12 | 46.7 | |
| Video-CCAM-14BFrames=962026.04 | 46.7 | |
| Kangaroo-7BFrames=642026.04 | 46.6 | |
| LongVA-7BFrames=1282026.04 | 46.2 | |
| VideoChat-Online-8BFrames=2fps2026.04 | 44.9 | |
| VideoLLaMA2Size=7B, Input Frames=>=642025.12 | 43.8 | |
| Video-LLaVASize=7B, Input Frames=>=642025.12 | 38.1 | |
| ShareGPT4VideoSize=8B, Input Frames=>=642025.12 | 37.9 | |
| MovieChat-7BFrames=20482026.04 | 33.4 | |
| VideoChat2-7BFrames=162026.04 | 33.2 |