Tongyi DeepResearch Launches! Fully Open-Source AI Model Makes Research Simpler

In the field of artificial intelligence, the latest research results released by the Tongyi DeepResearch team have attracted widespread attention. This breakthrough not only enables AI to "chat" but also elevates it to "conduct research." Moreover, it demonstrates its outstanding performance to the world with an open attitude. Tongyi DeepResearch has achieved state-of-the-art results in multiple authoritative benchmark tests, and its overall capabilities even exceed many internationally renowned models. Furthermore, the model, framework, and solutions are fully open-sourced, truly bringing the productivity of deep research to everyone.

Compared to expensive and restricted models internationally, the Tongyi DeepResearch team has chosen a completely open approach, providing a series of tools and solutions. In multiple test projects, such as Humanity's Last Exam, BrowseComp, GAIA, etc., the lightweight model 30B-A3B performed excellently, achieving state-of-the-art levels.

Notably, the team shared detailed methodologies for building the DeepResearch Agent on their official website and GitHub, covering the entire process from data synthesis to reinforcement learning. In terms of reasoning, the model demonstrates two advantages: the basic ReAct mode can fully unleash the model's potential without complex prompts, while the deep mode further enhances its performance in complex reasoning and planning capabilities.

The Tongyi DeepResearch team has also made significant contributions to data strategies. They adopted a multi-stage data strategy, generating high-quality training data through incremental pre-training and post-training methods without relying on expensive manual annotations. Additionally, the model's reasoning modes are divided into two types: the native ReAct mode and the Heavy mode for complex tasks, which provides convenience for solving highly challenging research tasks.

During the reinforcement learning process, the Tongyi DeepResearch team continuously improved the model's performance by optimizing algorithms and stable infrastructure. They adopted targeted strategies to ensure precise signal matching during the learning process, ultimately achieving efficient learning outcomes.

The release of Tongyi DeepResearch not only provides a new direction for AI research but also responds positively to the global technology community, demonstrating the strong power of open collaboration.