Atlas

Benchmarks

← All benchmarks

Video-MME - Long Videos with Subtitles

Multimodal · 2024-05-31

The long-duration subset of Video-MME (videos of roughly 30-60 minutes, 900 multiple-choice questions) answered with subtitles provided, as OpenAI reports it under "VideoMME (long, with subtitle category)". Scores are answer accuracy; not comparable with the overall no-subtitles Video-MME score. Higher is better.

Top models (higher is better)

ModelScore
GPT-586.7
o384.9
O4 Mini79.5
GPT-4.178.7
GPT-5 Mini78.5
GPT-4.1 Mini68.4
GPT-5 Nano65.7
GPT-4.1 Nano55.2
Loading Atlas data…