Deep Research Max tier

Google has split its Gemini research agent into two tiers, adding a new Deep Research Max version for longer, heavier jobs. (blog.google) Google announced Deep Research and Deep Research Max on April 21, 2026, and said both run on Gemini 3.1 Pro, the company’s latest reasoning model released in February. Deep Research is positioned as the faster, lower-latency option, while Max is built for “maximum comprehensiveness” and background work. (blog.google 1) (blog.google 2) The product is API-only in preview and works through Google’s Interactions API rather than the standard `generate_content` endpoint. Google’s developer docs list Deep Research Max as `deep-research-max-preview-04-2026` and say research jobs can take several minutes, so developers must run them asynchronously in the background. (ai.google.dev) Google said Deep Research Max scored 93.3% on DeepSearchQA, a benchmark Google introduced in December for testing how well agents handle broad web research. The company said the Max tier uses more “test-time compute,” meaning it spends more steps searching, reading and refining before it returns a report. (blog.google 1) (blog.google 2) That setup pushes the product toward analyst-style workflows instead of chatbot replies. Google’s launch post uses a nightly due-diligence report for an analyst team as its example, with the agent searching the web, internal files and outside data services before morning. (blog.google) Google is also trying to make the agent useful inside companies that do not want research limited to public webpages. The new version can pull from uploaded files, connected file stores and external tools through Model Context Protocol, a standard for plugging AI systems into other software and data sources. (blog.google) (ai.google.dev) The pricing points to that same audience. Google’s developer docs estimate a typical Deep Research Max task at about $1 to $3, based on model tokens and search queries, while the broader Gemini 3.1 Pro pricing page lists paid rates for input, output and grounded Google Search usage. (ai.google.dev 1) (ai.google.dev 2) Google is not the only company selling “deep research” agents. OpenAI expanded its own Deep Research product to ChatGPT Plus, Team, Education and Enterprise users in February 2025, and Google has also been adding deep-search style features inside Search and the Gemini app. (venturebeat.com) (blog.google) (blog.google) For now, Google is drawing a line between a research assistant that answers quickly and one that keeps working after the user leaves. Deep Research Max is the one aimed at teams willing to trade time and money for a thicker report by morning. (blog.google)