Google launches Deep Research Max specs

Google on April 21 rolled out Deep Research Max in preview through the Gemini API, a new version of its autonomous research agent built for longer jobs. (ai.google.dev) (blog.google) A research agent is software that plans a topic, searches for sources, reads them, and writes a report instead of answering in one shot. Google said the Max version is aimed at “deep competitive landscape analysis” and due diligence. (ai.google.dev) (blog.google) Google’s documentation says Deep Research Max can use up to about 160 search queries, about 900,000 input tokens, and about 80,000 output tokens in a single run. The company labels it “maximum comprehensiveness” and lists the model name as `deep-research-max-preview-04-2026`. (ai.google.dev 1) (ai.google.dev 2) The same April 21 release added collaborative planning, which lets a user review the research plan before the agent starts. Google also added visualization support, Model Context Protocol server integration, and File Search. (ai.google.dev) (blog.google) Google said the standard Deep Research preview is tuned for speed and efficiency, while Deep Research Max is tuned for breadth across hundreds of sources. That splits the product into a faster agent for interactive use and a heavier agent for longer research workflows. (ai.google.dev 1) (ai.google.dev 2) This launch extends work Google started in December 2025, when it brought the Deep Research agent to developers through the Interactions API. In that earlier release, Google also introduced DeepSearchQA, an open benchmark for testing how complete web-research agents are. (blog.google) Google’s consumer Gemini app had already offered Deep Research, including links to sources and, later, Workspace app connections. The April 2026 release moves the product further into developer tooling, where agents can be embedded inside custom applications. (blog.google 1) (blog.google 2) (blog.google 3) Google’s broader Gemini API pitch is that developers can feed in very large files and long context windows, then connect models to tools and external systems. Deep Research Max packages those pieces into a single agent that plans, searches, reads, and synthesizes on its own. (ai.google.dev) (ai.google.dev) For now, Google is calling Deep Research Max a preview, which means the specs and limits can still change. But the April 21 release makes clear where Google is pushing next: bigger, slower, more autonomous research jobs for developers who need more than a chatbot answer. (ai.google.dev) (ai.google.dev)