On July 15, Mistral AI Search announced the completion of a new iteration of its "Deep Research" module and its official public beta launch, becoming the first deep research-level search service in China that is freely available to the public and features multi-turn reasoning chain visualization.
The upgraded system adopts a segmented reinforcement learning strategy, breaking down the originally resource-intensive "Deep Research" tasks into multiple sub-tasks. This approach maintains result accuracy while reducing operational costs to a level that allows for free public access, with particularly strong performance in retrieving and reasoning about Chinese text.
Users simply need to switch to the "Deep Research" mode on Mistral's homepage, input complex questions, and receive a comprehensive report that unfolds through a "question chain," automatically retrieves information, cross-validates it, and presents it in tables or paragraphs. The entire reasoning process can be viewed and traced in real time.
Official benchmark comparisons show that this version outperforms the latest publicly available models such as Tongyi WebSailor in Chinese open QA, fact verification, and long-chain reasoning tasks.
Mistral stated that over the past five months, the team has optimized from four dimensions: data cleaning, index structure, retrieval algorithms, and reasoning scheduling. The goal is to make advanced research capabilities accessible to all, allowing researchers, students, and small businesses without budgets for expensive subscriptions to instantly gain professional insights with one click.