Long-video understanding on LVBench 4101 sec
48.2ScoreVideoChat-Flash
Evaluation results
| Method | Links | |
|---|---|---|
| VideoChat-FlashSize=7B, Frames=5122026.06 | 48.2 | |
| CoVERSize=7B, Frames=768L + 32H, Context=16k2026.06 | 46.6 | |
| Hour-LLaVASize=7B, Frames=1fps2026.06 | 45.6 | |
| VILAMPSize=7B, Frames=1fps2026.06 | 45.2 | |
| VambaSize=10B, Frames=10242026.06 | 42.1 | |
| Qwen2.5-VL*Size=7B, Frames=128, Context=16k2026.06 | 42 | |
| VideoMindSize=7B2026.06 | 40.8 | |
| ConanSize=7B, Frames=322026.06 | 39.2 | |
| GPT-4oFrames=1fps2026.06 | 34.7 | |
| LLaMA-VIDSize=7B, Frames=1fps2026.06 | 23.9 |