Long-video understanding on LongVideoBench (Duration-Specific Metrics)
66.7Average PerformanceGPT-4o
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GPT-4oFrames=-, LLM Size=-2026.06 | 66.7 | — | |
| MiMo-VL w/ Q-FoldFrames=32, LLM Size=7B2026.06 | 64.1 | 59.6 | |
| Gemini-1.5-ProFrames=-, LLM Size=-2026.06 | 64 | — | |
| Qwen2.5-VL w/ Q-FoldFrames=32, LLM Size=7B2026.06 | 63 | 56.7 | |
| Qwen2-VL w/ FOCUSFrames=32, LLM Size=7B2026.06 | 62.3 | — | |
| Gemini-1.5-FlashFrames=-, LLM Size=-2026.06 | 61.6 | — | |
| GPT-4VFrames=-, LLM Size=-2026.06 | 61.3 | — | |
| Qwen2-VL w/ Q-FoldFrames=32, LLM Size=7B2026.06 | 60.7 | 55.1 | |
| Qwen2-VL w/ AKSFrames=32, LLM Size=7B2026.06 | 60.5 | — | |
| LLaVA-OneVision-1.5 w/ Q-FoldFrames=32, LLM Size=8B2026.06 | 60.1 | 55.5 | |
| mPLUG-Owl3Frames=128, LLM Size=7B2026.06 | 59.7 | — | |
| MiMo-VLFrames=32, LLM Size=7B2026.06 | 59.1 | 50.2 | |
| Qwen2-VL w/ GenS†Frames=54/50, LLM Size=7B2026.06 | 58.7 | — | |
| Qwen2-VL w/ TSPO†Frames=32, LLM Size=7B2026.06 | 58.6 | — | |
| Qwen2.5-VLFrames=32, LLM Size=7B2026.06 | 58.5 | 49.6 | |
| Qwen2-VL w/ Q-FrameFrames=4 + 8 + 32, LLM Size=7B2026.06 | 58.4 | 52.8 | |
| LongVILAFrames=128, LLM Size=7B2026.06 | 57.1 | — | |
| LLaVA-OneVision-1.5Frames=32, LLM Size=8B2026.06 | 56.6 | 47.9 | |
| Oryx-1.5Frames=64, LLM Size=7B2026.06 | 56.3 | — | |
| TimeMarkerFrames=128, LLM Size=8B2026.06 | 56.3 | — | |
| VAMBAFrames=1024, LLM Size=13B2026.06 | 55.9 | — | |
| Qwen2-VLFrames=32, LLM Size=7B2026.06 | 55.5 | 47.9 | |
| Video-XLFrames=128, LLM Size=7B2026.06 | 50.7 | — | |
| LongVAFrames=128, LLM Size=7B2026.06 | 47.8 | — |