WideSeek: Advancing Wide Research via Multi-Agent Scaling

WideSeek targets Wide Research: retrieving and synthesizing comprehensive information sets in parallel under complex constraints. We introduce (1) WideSeekBench, a General Broad Information Seeking (GBIS) benchmark focused on search breadth, and (2) WideSeek, a dynamic hierarchical multi-agent system optimized with an end-to-end RL framework.

Highlights

• WideSeekBench: a benchmark built via a rigorous multi-stage data pipeline to ensure diversity across domains, target set sizes, and logical constraints.
• WideSeek system: a dynamic hierarchical multi-agent architecture that can autonomously fork parallel sub-agents on demand.
• Unified RL training: a framework that linearizes multi-agent trajectories and optimizes the whole system end-to-end with reinforcement learning.

Experiments

• WideSeekBench:
• Browsecomp-Plus:

Code

We will release the code soon.

Citation

If you find this work useful, please cite:

@misc{huang2026wideseekadvancingwideresearch,
      title={WideSeek: Advancing Wide Research via Multi-Agent Scaling}, 
      author={Ziyang Huang and Haolin Ren and Xiaowei Yuan and Jiawei Wang and Zhongtao Jiang and Kun Xu and Shizhu He and Jun Zhao and Kang Liu},
      year={2026},
      eprint={2602.02636},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2602.02636}, 
}

License

MIT

Contact

Feel free to contact Ziyang via email: huangzy0312@gmail.com