Skip to content

Video-MME

A benchmark for video analysis by multimodal models

Visit official site

What Video-MME is

The benchmark is described as a platform for evaluating multimodal large language models on video analysis, filling a gap in how continuous visual data is measured.

What you can do with it

  • Read accuracy scores by video length
  • Compare models across 6 domains and 30 subcategories
  • Review statistics on duration and task type distribution

Getting started

  • Visit the official site and review the evaluation criteria
  • Choose an MLLM to test
  • Submit the model and read per-subcategory results

Pros & cons

✓ What we like

  • Newly collected data with human annotation
  • Coverage of video length and task type is described as comprehensive

! What to watch out for

  • The described use is model performance testing
  • Results are organised by video subcategory

FAQ

How many domains are covered?

The material states 6 main domains and 30 subcategories.

Where does the data come from?

It is described as newly collected and human-annotated, not an existing dataset.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review