Long-Video Understanding on LVBench 30m~2h
56.6ScoreQwen2.5+FlexSelect
Evaluation results
| Method | Links | |
|---|---|---|
| Qwen2.5+FlexSelectModel Type=Transformer, Size=72B2026.02 | 56.6 | |
| Nemotron-Nano-V2 VL + All layer reductionModel Type=Hybrid, Size=12B2026.02 | 54.29 | |
| LLaVA-Video+FlexSelectModel Type=Transformer, Size=7B2026.02 | 52.9 | |
| Qwen2.5 VL+FlexSelectModel Type=Transformer, Size=7B2026.02 | 51.2 | |
| Qwen3 VL + All layer reductionModel Type=Transformer, Size=8B2026.02 | 45.45 | |
| VideoLLaMA3Model Type=Transformer, Size=7B2026.02 | 45.3 | |
| SF-LLaVA-1.5Model Type=Transformer, Size=7B2026.02 | 45.3 | |
| ViLAMPModel Type=Transformer, Size=7B2026.02 | 45.2 | |
| Qwen2-VLModel Type=Transformer, Size=7B2026.02 | 42.4 | |
| GPT-4oModel Type=Transformer, Size=–2026.02 | 34.7 | |
| Gemini-1.5-ProModel Type=Transformer, Size=–2026.02 | 33.1 | |
| Oryx-1.5Model Type=Transformer, Size=34B2026.02 | 30.8 |