Tongyi DeepResearch Technical Report
Tongyi DeepResearch Team, Baixuan Li, Bo Zhang, Dingchu Zhang, Fei Huang, Guangyu Li, Guoxin Chen, Huifeng Yin, Jialong Wu, Jingren Zhou, Kuan Li, Liangcai Su, Litu Ou, Liwen Zhang, Pengjun Xie, Rui Ye, Wenbiao Yin, Xinmiao Yu, Xinyu Wang, Xixi Wu, Xuanzhong Chen, Yida Zhao
2025-10-29
Summary
This paper introduces Tongyi DeepResearch, a new large language model designed to perform complex research tasks that require digging for information over a long period of time.
What's the problem?
Current large language models often struggle with research projects that aren't straightforward. They have trouble staying focused on a long-term goal, finding relevant information from multiple sources, and consistently reasoning through complex problems. Creating models that *can* do this effectively usually requires a lot of expensive human effort to label data and guide the model's learning.
What's the solution?
The researchers created Tongyi DeepResearch using a special training process that happens in two phases: during training and after training. This process encourages the model to act like a research agent, independently seeking out information. Crucially, they developed a way to automatically create the training data needed, avoiding the need for costly human labeling. They also built simulated environments to help the model learn to interact consistently and reliably. The model itself has a large number of parameters, but cleverly only uses a smaller subset at a time, making it more efficient.
Why it matters?
Tongyi DeepResearch represents a significant step forward in AI's ability to conduct in-depth research autonomously. It performs better than existing models on several challenging research benchmarks, and by releasing the model and the tools used to build it, the researchers hope to encourage further development in this area, potentially leading to AI systems that can assist with complex problem-solving and knowledge discovery.
Abstract
We present Tongyi DeepResearch, an agentic large language model, which is specifically designed for long-horizon, deep information-seeking research tasks. To incentivize autonomous deep research agency, Tongyi DeepResearch is developed through an end-to-end training framework that combines agentic mid-training and agentic post-training, enabling scalable reasoning and information seeking across complex tasks. We design a highly scalable data synthesis pipeline that is fully automatic, without relying on costly human annotation, and empowers all training stages. By constructing customized environments for each stage, our system enables stable and consistent interactions throughout. Tongyi DeepResearch, featuring 30.5 billion total parameters, with only 3.3 billion activated per token, achieves state-of-the-art performance across a range of agentic deep research benchmarks, including Humanity's Last Exam, BrowseComp, BrowseComp-ZH, WebWalkerQA, xbench-DeepSearch, FRAMES and xbench-DeepSearch-2510. We open-source the model, framework, and complete solutions to empower the community.