Google just launched Deep Research and Deep Research Max — two autonomous AI research agents running on Gemini 3.1 Pro. And the numbers suggest Google isn’t holding back.
What Deep Research Max does
Deep Research Max autonomously plans research, searches the web (averaging 160 searches per session), reads your private documents, analyzes the results, and delivers a complete report — with source citations and native charts. Essentially what a research analyst does, just significantly faster.
On the DeepSearchQA benchmark, the system scores 93.3 percent — a massive jump from 66.1 percent in December 2025. That’s not an incremental improvement, that’s a generational leap.
Enterprise integration from day one
What makes this particularly interesting: Deep Research Max supports the Model Context Protocol (MCP) for connecting to enterprise data. At launch, there are integrations with FactSet, S&P Global, and PitchBook — exactly the data sources analysts at banks and consulting firms use daily.
A typical session costs around $4.80 — not cheap, but when you consider what a human analyst charges per hour, it’s a different conversation.
The bigger picture
Deep Research Max is Google’s answer to the agentic AI wave. While Anthropic pushes desktop automation with Cowork and Claude Code, and OpenAI expands Codex into an everything-tool, Google is doubling down on the research use case.
That’s a smart bet, because research is one of the areas where AI already delivers real value today. No ‘vibe coding’, no ‘control my computer’ — just: read 200 sources, find the relevant facts, and write me a report with charts. That’s a concrete problem, and Google seems to be solving it much better now than four months ago.
The MCP support is the really interesting part. It opens Deep Research Max to arbitrary data sources — and that makes it relevant for companies that want to combine their own data with public sources.
Sources: