OVBench: leaderboard
Online video-understanding QA benchmark measuring a model's ability to perceive, memorize, and reason over streaming video across past, current, and future temporal contexts — 6 core task types forming 16 subtasks.
Metric: AVG. Source: huggingface.co. Saturation forecast: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 1.5 Flash | 50.7 | #618 |
| 2 | Qwen 2 VL 7B | 49.7 | #750 |
| 3 | MiniCPM-V-2.6 | 39.1 | #836 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=ovbench · How It Works · Data refreshed daily, snapshot 2026-10-20.