Alibaba joined the Open-Source AI agent race with the launch of Tongyi DeepResearch Agent. On September 18, 2025, the Chinese tech giant revealed its leading open-source deep research AI system, which rivals OpenAI’s Deep Research platform.
Released under the permissive Apache 2.0 license, the Alibaba model will be made available through HuggingFace, GitHub, and ModelScope to ensure accessibility for developers, enterprises, and researchers. The company claimed its 30-billion-parameter model is as capable as OpenAI’s Deep Research tool and is more efficient.
Reportedly, the Tongyi DeepResearch Agent is powered by Qwen3-30B-A3B, a foundation model that features nearly 30 billion parameters, and around 3 billion of them are activated during inference.
Although smaller in size, this open-source model has shown impressive performance in academic and other practical fields. As reported, it achieved 32.9% accuracy on Humanity's Last Exam, compared to OpenAI's Deep Research, which achieved a score of 26.6%. In addition, it scored 43.4% on BrowseComp in English, 46.7% in Chinese, and 75% on xbench-DeepSearch.
These tests demonstrate its advantage in legal research, where it produces more genuine results and citations than other similar AIs.
Alibaba has already integrated Tongyi into real-world applications, such as the company’s navigation platform, Amap. Here, the AI model helps users to plan their multi-day travel itineraries.
In the legal tech space, Alibaba has adapted it into Tongyi FaRui, which allows users to discover Chinese case law with trustworthy citations. This indicates that the tech giant plans to integrate the technology into both consumer and enterprise products, beyond its current research use.
The key question is how Alibaba’s legal research technology will stand against OpenAI's dominance. OpenAI's models, such as GPT-4, utilize massive compute resources, enormous training datasets, and significant third-party integrations. Alibaba, however, emphasizes efficiency, which means the latest Alibaba model can rival other heavyweight opponents without similar infrastructure.
What makes Tongyi truly different from others is its open-source status. While OpenAI models lack transparency and customizability, Tongyi will be best for those regions where trust, localization, and adaptability are priorities.
Also Read: Nano Banana Brings a Breakthrough, Google Gemini Beats OpenAI’s ChatGPT in Downloads
Alibaba’s move is more than just the launch of an AI model. It, instead, indicates a strategic shift in the global AI competition. By focusing on openness, parameter efficiency, and real-world usability, the Alibaba model challenges the dominance of other established Deep Research agents.
Tongyi’s launch marks a major move in the AI race. It has deviated from major models in several aspects, including model size, efficiency, and deployment strategy. If Alibaba maintains the momentum, Tongyi could lead the industry toward a more transparent AI future.