LVOmniBench: leaderboard
Metric: Accuracy (%) on the 1,014 four-option multiple-choice questions about 275 long open-domain videos (10 to 90 minutes, mean 34.5 minutes) in LVOmniBench, each model at its maximum permitted number of input frames (768 for the Qwen models and video-SALMONN 2+, 64 for MiniCPM-o 4.5, 16 for VideoLLaMA2, 1 frame per second for Gemini with thinking), with the video's frames and its audio track as input; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 65.8 | #77 |
| 2 | Gemini 3 Flash | 59 | #93 |
| 3 | Gemini 2.0 Flash | 42.9 | #331 |
| 4 | Qwen2.5-Omni-7B | 32 | #601 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=lvomnibench · How It Works · Data refreshed daily, snapshot 2026-10-11.