OmniVideo-100K is an audio and video inference dataset for fine-tuning instructions in multimodal large language models, released in June 2026 by Nanjing University in collaboration with the Institute of Automation, Chinese Academy of Sciences. Related research papers include... OmniVideo-100K: A Dataset for Audio-Visual Reasoning through Structured Scripts and Evidence ChainsIt aims to enhance the collaborative capabilities of models in long-term time series and cross-modal dependencies, and is widely used to train and evaluate the capabilities of multimodal large models in video understanding and complex reasoning tasks. This dataset contains 100,000 question-answer pairs from 5,214 YouTube videos, divided into 70,000 open-ended question-answer pairs and 30,000 multiple-choice questions based on task format. It covers 10 types of audio and video question-answering tasks and is divided into three cognitive levels: alignment, understanding, and reasoning.

Citation

@article{cai2026omnivideo100k,
title={OmniVideo-100K: A Dataset for Audio-Visual Reasoning through Structured Scripts and Evidence Chains},
author={Cai, Xinyue and Fu, Chaoyou and Zhang, Yi-Fan and He, Ran and Shan, Caifeng},
journal={arXiv preprint arXiv:2606.14702},
year={2026}
}