Google has recently announced the launch of two new self-research agents based on the Gemini 3.1 Pro architecture—Deep Research and Deep Research Max. These are now available for public preview through the Gemini API paid tier. The tools aim to fully automate complex research processes, marking a shift in AI agents from simple web searches to "long-term computing" models with deep reasoning capabilities.
The standard version of Deep Research focuses on efficiency and low latency, suitable for real-time conversation scenarios requiring immediate feedback; while Deep Research Max prioritizes research depth, using extended computation time for multi-round reasoning and iteration, mainly targeting asynchronous back-end tasks such as due diligence reports. In terms of technical implementation, the new version introduces support for the Model Context Protocol (MCP), allowing agents to access open network information as well as proprietary databases such as finance or market data. Additionally, the agents now have native visualization capabilities, enabling direct generation of HTML-formatted charts and infographics.
In performance benchmark tests, Google claims that Deep Research Max shows significant improvements in retrieval and reasoning tasks compared to its predecessor. However, industry experts point out that the comparison results with OpenAI GPT-5.4 series and Anthropic Opus4.6 are influenced by testing methods and should be interpreted with caution. Notably, this series of agents has added a collaboration planning feature, supporting multi-modal inputs such as PDFs, audio, and video, and allows developers to completely disable network access to ensure private data security.
Google stated that these two agents share the same research architecture as NotebookLM and Google Search, and will be further introduced to the enterprise market through Google Cloud. As self-research agents enter the "long-term reasoning" phase, the role of AI in professional analysis will transition from an information transporter to a deep analysis expert with autonomous planning capabilities.
