Kwai-Keye/VideoTemp-Bench

VideoTemp-o3: Harmonizing Temporal Grounding and Video Understanding in Agentic Thinking-with-Videos

Illustration of the agentic pipeline in VideoTemp-o3. Given a video QA pair, the model performs on-demand temporal grounding to locate the most relevant segment, then refines it iteratively. Finally, it produces a reliable answer grounded in the pertinent visual evidence.

Data Source

Dataset	Source
NExT-GQA	https://github.com/doc-doc/NExT-GQA
LongVideo-Reason	https://github.com/NVlabs/Long-RL
LongVILA	https://huggingface.co/datasets/LongVILA/longvila_sft_dataset
ScaleLong	https://huggingface.co/datasets/m-a-p/ScaleLong

Citation

If you find our work useful, please consider citing:

@article{liu2026videotemp,
  title={VideoTemp-o3: Harmonizing Temporal Grounding and Video Understanding in Agentic Thinking-with-Videos},
  author={Liu, Wenqi and Wang, Yunxiao and Ma, Shijie and Liu, Meng and Su, Qile and Zhang, Tianke and Fan, Haonan and Liu, Changyi and Jiang, Kaiyu and Chen, Jiankang and Tang, Kaiyu and Wen, Bin and Yang, Fan and Gao, Tingting and Li, Han and Wei, Yinwei and Song, Xuemeng},
  journal={arXiv preprint arXiv:2602.07801},
  year={2026}
}