Long Video Understanding on VideoMME (test)
64.2AccuracyNVILA-8B
Evaluation Results
| Method | Links | |
|---|---|---|
| NVILA-8BBackbone=8B2026.05 | 64.2 | |
| LLaVA-Video-7B + ST-GridPoolInput frames=64, Pooling Strategy=ST-GridPool2026.05 | 64.2 | |
| LLaVA-Video-7BInput frames=642026.05 | 63.3 | |
| Apollo-7BBackbone=7B2026.05 | 61.3 | |
| LLaVA-OneVision-7B + ST-GridPoolInput frames=32, Pooling Strategy=ST-GridPool2026.05 | 59 | |
| Oryx-1.5-7BBackbone=1.5B2026.05 | 58.8 | |
| LLaVA-OneVision-7BInput frames=322026.05 | 58.2 | |
| IXC-2.5-7BBackbone=7B2026.05 | 55.8 | |
| VideoLLaMA2.1-7BBackbone=7B2026.05 | 54.9 | |
| mPLUG-Owl3-8BBackbone=8B2026.05 | 53.5 | |
| LongVA-7BBackbone=7B2026.05 | 52.6 |