Google DeepMind · 2026-04-21 · major
Google Deep Research Max — Autonomous Research Agent Hits 93.3% DeepSearchQA with MCP Support
Google launched Deep Research and Deep Research Max in the Gemini API — autonomous multi-step research agents powered by Gemini 3.1 Pro. Deep Research Max scores 93.3% on DeepSearchQA (up from 66.1%) and 54.6% on Humanity's Last Exam. Both add MCP support for private data sources and native chart generation.
Google's Deep Research Max is an async research agent that plans, iterates, and synthesizes — now with MCP access to private data and native chart generation.
Key specs
| Deep search qa (max) | 93.3% |
|---|---|
| Previous deep search qa | 66.1% |
| Humanity's last exam (max) | 54.6% |
| Underlying model | Gemini 3.1 Pro |
What is it?
Google released two new variants of its Deep Research agent in the Gemini API: Deep Research (lower latency, for interactive UIs) and Deep Research Max (higher quality, async, for exhaustive background runs). Both run on Gemini 3.1 Pro. New in this release: arbitrary MCP server connections for private enterprise data, native in-line chart and infographic generation, collaborative planning (review and edit the research plan before execution), and full multimodal input support (PDFs, CSVs, images, audio, video).
How does it work?
Deep Research Max uses extended test-time compute — it iteratively plans, searches, and refines its analysis over multiple passes before producing a final cited report. MCP server support lets it reach financial data providers, internal databases, or any MCP-compatible endpoint mid-session, not just the public web. Intermediate reasoning steps stream in real time. The chart generation capability uses either HTML or Google's Nano Banana format to produce presentation-ready visualizations inline with the report.
Why does it matter?
The 27-point jump on DeepSearchQA (66.1% → 93.3%) is a meaningful quality step. More practically, MCP support transforms Deep Research from a web-only tool into one that can reason over proprietary data, making it directly competitive for internal research workflows. The async Max mode pairs well with cron-based pipelines that need nightly due-diligence reports or ongoing competitive analysis.
Who is it for?
Enterprise developers building research automation; analysts who need deep synthesis across private and public data sources.
Try it
model='gemini-deep-research-max-04-2026' via Gemini API (public preview, paid tiers)