Google DeepMind launches Gemini Deep Research Max

Google DeepMind introduced a two-tier successor to its Deep Research agent, built on Gemini 3.1 Pro: a faster, lower-cost Deep Research mode for interactive use, and Deep Research Max, which trades speed for extended test-time compute and, Google said, more thorough sourcing and analysis. Both replaced the preview research agent Google had shipped in December 2025.

The agents added support for the Model Context Protocol, letting them pull from a user’s own data sources alongside the open web, and could take manual uploads of spreadsheets, PDFs, images, audio and video. Reports could include native charts and infographics rather than text summaries alone, and users could review and edit a research plan before the agent executed it. Google said Deep Research Max consulted substantially more sources than the December version and produced improved factuality on internal benchmarks, without publishing comparative scores.

Coverage described the split as aimed at different jobs: Deep Research for quick, interactive queries such as a financial dashboard answering an analytical question in near real time, and Deep Research Max for longer asynchronous workflows such as a literature review or investment memo. Both launched in public preview through Google’s paid API tiers, with Google saying availability for startups and enterprise customers via Google Cloud would follow. The release extended a pattern among the frontier labs of packaging a general model as a dedicated, slower “deep research” product, following OpenAI’s and xAI’s earlier versions of the same idea.