Long Video Understanding on LVBench (30-90 min)
69.2AccuracyGemini-2.5-Pro
Evaluation results
| Method | Links | |
|---|---|---|
| Gemini-2.5-ProInput Frames=>=642025.12 | 69.2 | |
| Seed1.5-VLSize=8B, Input Frames=>=642025.12 | 64.6 | |
| GPT-4oInput Frames=>=642025.12 | 64.4 | |
| VideoMem*Size=8B, Input Frames=642025.12 | 58.5 | |
| Qwen3-VL*Size=8B, Input Frames=642025.12 | 50.6 | |
| Video-XL2Size=8B, Input Frames=>=642025.12 | 48.4 | |
| Qwen2.5-VLSize=72B, Input Frames=>=642025.12 | 47.3 | |
| Qwen2.5-VLSize=7B, Input Frames=>=642025.12 | 45.3 | |
| VideoLLaMA3Size=7B, Input Frames=>=642025.12 | 45.3 | |
| NVILASize=8B, Input Frames=>=642025.12 | 44 | |
| LLaVA-videoSize=7B, Input Frames=>=642025.12 | 41.5 | |
| LongVASize=7B, Input Frames=>=642025.12 | 36.2 | |
| LLaVA-OnevisionSize=7B, Input Frames=>=642025.12 | 26.9 |