Google DeepMind's Deep Research agents run 160 searches while you sleep

Google DeepMind has released two new autonomous research agents—Deep Research and Deep Research Max—in public preview via the Gemini API. Built on Gemini 3.1 Pro, the standard Deep Research agent prioritizes speed and low cost, designed for interactive user-facing products. Deep Research Max is the heavyweight option, running asynchronous background workflows such as overnight due diligence reports. It uses extended test-time compute to iteratively search and refine, executing roughly 160 search queries per task. Benchmarks show 93.3% on DeepSearchQA for web research and 85.9% on BrowseComp for hard-fact retrieval.

The agents can search the open web, user file uploads, and connected data sources through Model Context Protocol (MCP) servers. They generate charts and infographics natively and consult over 100 sources in a single task. Inputs include PDFs, CSVs, images, audio, and video. Users can review and edit the research plan before execution, with real-time reasoning streaming. Google has partnered with FactSet, S&P Global, and PitchBook for MCP integrations, positioning Deep Research as an enterprise workflow engine for finance and life sciences. The infrastructure already powers research features in Gemini App, NotebookLM, Google Search, and Google Finance, and is now exposed to developers via the Interactions API and Google AI Studio.