Google Launches Deep Research Max in Gemini

What changed

Google released Deep Research and Deep Research Max on April 21, 2026, built on Gemini 3.1 Pro. Both improve on the December 2025 preview with lower latency, reduced cost, and higher quality scores. New capabilities: Model Context Protocol (MCP) support for custom data streams, native charts and infographics in HTML or Nano Banana format, collaborative research planning, extended tooling like Google Search and code execution, multimodal inputs spanning PDFs, CSVs, images, audio, and video, and real-time streaming of reasoning steps. Deep Research prioritises speed for interactive apps; Deep Research Max uses extended compute for offline reports and posts higher retrieval and reasoning numbers.

Why it matters

This turns research into the entry point for agentic pipelines, mixing web data with proprietary sources via MCP. Google is explicitly targeting finance, life sciences, and market research, where pros need gated data handled with audit trails. The threat envelope covers standalone summarisers and manual analyst workflows: one API call returns cited, visual reports with the reasoning trail attached. The same substrate already powers Gemini App, NotebookLM, Search, and Finance, so the engineering bar is enterprise-grade rather than demo-grade.

What to watch for

Watch confirmed MCP integrations from FactSet, S&P Global, and PitchBook: those validate the regulated-data play. Persistent high cost or hallucination rates on nuanced topics break the thesis. General availability is expected by Q3 2026, with more MCP tools for regulated data likely between now and then. Pull the docs at https://ai.google.dev/gemini-api/docs/deep-research, run Max asynchronously against a real internal workload, and benchmark cost per useful report against your current retrieval stack before committing production traffic.