Deep Research Max: a step change for autonomous research agents
Key Points
- Gemini 3.1 Pro powers Deep Research Max
- MCP support for proprietary data integration
- Native charts plus iterative, high‑quality synthesis
Summary
Deep Research Max (powered by Gemini 3.1 Pro) is an upgraded autonomous research agent designed for exhaustive, high‑quality, long‑horizon research workflows. It combines extended test‑time compute, web + proprietary data access via MCP, native visualizations, and iterative reasoning to produce fully cited, presentation‑ready reports for domains like finance and life sciences.
Key Points
- Two agent tiers: Deep Research (low-latency, cost-optimized for interactive surfaces) and Deep Research Max (max comprehensiveness for asynchronous/background workflows).
- Powered by Gemini 3.1 Pro and extended test‑time compute to iteratively reason, search, and refine outputs for higher factuality and synthesis quality.
- Model Context Protocol (MCP) support enables secure integration of proprietary data streams and custom MCP servers alongside web sources.
- Native visual output: generates charts and infographics inline (HTML / Nano Banana) for presentation‑ready reports.
- Extended tooling and controls: combine Google Search, MCP, URL Context, Code Execution, File Search; turn web access on/off; collaborative planning to review/adjust plans before execution.
- Multimodal grounding: accept PDFs, CSVs, images, audio, video and file uploads as research inputs; real‑time streaming exposes intermediate reasoning and partial outputs.
- Availability: public preview today via paid tiers in the Gemini Interactions API; coming to select Google Cloud offerings and partner MCP integrations.
Practical notes for engineers
- Trigger exhaustive workflows with a single API call; choose scope (web, MCPs, files) to control data sources and compliance.
- Use collaborative planning to pin research scope and reduce hallucination risk before execution.
- Combine Code Execution and File Search to run data transforms and produce native charts automatically.
- Recommended for nightly/offline batch jobs (Deep Research Max) or interactive UIs (Deep Research) depending on latency and cost constraints.
Where to start
See the Gemini Interactions API developer docs to configure Deep Research/Deep Research Max, register MCP servers, and enable multimodal inputs.