童译 DeepResearch 技术报告
计算与语言
2026-05-19 v3 人工智能
信息检索
机器学习
多智能体系统
摘要
我们 presented Tongyi DeepResearch，这是一种专为长时程、深层信息检索研究任务而设计的代理式大语言模型。为激励自主深层研究行为，Tongyi DeepResearch 通过一种端到端训练框架开发，结合了代理式中期训练和代理式后期训练，实现跨复杂任务的可扩展推理与信息检索。我们设计了高度可扩展的数据合成管道，完全自动化，无需依赖高成本的人类标注，赋能所有训练阶段。通过为每个阶段构建定制环境，我们的系统实现了稳定一致的交互。Tongyi DeepResearch 采用 305 亿总参数，其中仅 33 亿参数在每个 token 上被激活。在一系列代理式深层研究基准测试中实现了最先进的性能，包括 Humanity's Last Exam、BrowseComp、BrowseComp-ZH、WebWalkerQA、xbench-DeepSearch、FRAMES 和 xbench-DeepSearch-2510。我们开源了该模型、框架和完整解决方案，以期为社区提供支持。
引用
@article{arxiv.2510.24701,
title = {Tongyi DeepResearch Technical Report},
author = {Tongyi DeepResearch Team and Baixuan Li and Bo Zhang and Dingchu Zhang and Fei Huang and Guangyu Li and Guoxin Chen and Huifeng Yin and Jialong Wu and Jingren Zhou and Kuan Li and Liangcai Su and Litu Ou and Liwen Zhang and Pengjun Xie and Rui Ye and Wenbiao Yin and Xinmiao Yu and Xinyu Wang and Xixi Wu and Xuanzhong Chen and Yida Zhao and Zhen Zhang and Zhengwei Tao and Zhongwang Zhang and Zile Qiao and Chenxi Wang and Donglei Yu and Gang Fu and Haiyang Shen and Jiayin Yang and Jun Lin and Junkai Zhang and Kui Zeng and Li Yang and Hailong Yin and Maojia Song and Ming Yan and Minpeng Liao and Peng Xia and Qian Xiao and Rui Min and Ruixue Ding and Runnan Fang and Shaowei Chen and Shen Huang and Shihang Wang and Shihao Cai and Weizhou Shen and Xiaobin Wang and Xin Guan and Xinyu Geng and Yingcheng Shi and Yuning Wu and Zhuo Chen and Zijian Li and Yong Jiang},
journal= {arXiv preprint arXiv:2510.24701},
year = {2026}
}
备注
https://tongyi-agent.github.io/blog