Long Video Understanding on VideoMME Long w/o sub
70.3AccuracyLLaVA-Video
Evaluation results
| Method | Links | |
|---|---|---|
| LLaVA-VideoParam=72B, Frames=642026.03 | 70.3 | |
| Gemini-1.5-ProParam=-, Frames=2562026.03 | 67.4 | |
| VideoDetective (SeedVL-1.5)Param=20B(A), Frames=322026.03 | 65.6 | |
| GPT-4oParam=-, Frames=3842026.03 | 65.3 | |
| Qwen2.5-VLParam=72B, Frames=1282026.03 | 64.6 | |
| SeedVL-1.5Param=20B(A), Frames=322026.03 | 63.1 | |
| VideoDetective (Qwen3-VL)Param=8B, Frames=322026.03 | 55.6 | |
| LongVITA-16kParam=14B, Frames=642026.03 | 54.7 | |
| Qwen2.5-VLParam=7B, Frames=1282026.03 | 53.9 | |
| LongVILAParam=7B, Frames=1fps2026.03 | 53 | |
| LLaVA-VideoParam=7B, Frames=5122026.03 | 52.9 | |
| VideoXLParam=7B, Frames=1fps2026.03 | 52.3 | |
| InternVL-2.5Param=8B, Frames=322026.03 | 50.8 | |
| Qwen3-VLParam=8B, Frames=322026.03 | 50.2 | |
| VITA-1.5Param=7B, Frames=162026.03 | 47.1 | |
| LLaVA-OneVisionParam=7B, Frames=-2026.03 | 46.7 |