Google Releases Deep Research Max, an Agent for Hour-Long Analyses via API

On April 21, 2026, Google DeepMind released Deep Research Max, a new version of its autonomous research agent built on the Gemini 3.1 Pro model. Unlike the standard Deep Research, the Max variant is designed to prioritize depth of analysis over speed and is intended for asynchronous background tasks, such as overnight reports for analytics teams.

Two Modes, One Agent

Google split the product into two configurations. The standard Deep Research is meant to be optimized for speed and cost, aimed at interactive interfaces where the user waits for a live response. Deep Research Max spends significantly more compute on iterative reasoning and search, which Google says translates into broader source coverage and better synthesis of conflicting information in the final reports.

Both variants rely on the same research infrastructure that, according to Google, powers search and synthesis features in the Gemini app, NotebookLM, Google Search and Google Finance. This is the first time that infrastructure has been made available to developers as a standalone API product rather than a feature built into a specific consumer app.

Data Beyond the Open Web

A key new feature is support for the Model Context Protocol, which lets the agent securely connect not only to the open internet but also to a company's internal documents and specialized data sources, such as financial data. Google says FactSet, S&P Global and PitchBook are collaborating on MCP servers for market data, which is meant to open the door to use cases such as due diligence, competitive analysis and investment evaluation.

The agent also accepts input in multiple formats: PDFs, CSV spreadsheets, images, audio and video. Users can review and modify the agent's research plan in advance before it starts gathering data, and progress can be tracked live thanks to streaming of intermediate reasoning steps. There's also an option to restrict the agent to internal data only, with no web access.

Cost and Availability

Billing works on a pay-as-you-go model based on the base prices of the Gemini models. A task assigned to the standard Deep Research costs roughly $1-3, while a full Deep Research Max task can cost $3-7, depending on the complexity of the query and the number of searches performed. That's several times the cost of a single query to a regular chat model, which Google attributes to the scale of work the agent performs autonomously in the background.

For now, access is limited to developers using paid Gemini API tiers. Google says the agent will reach startups and large Google Cloud customers in the coming months, which is meant to expand use cases beyond the current tests in the finance and research sectors.

Competitive Backdrop

The launch is part of a broader race among major AI companies to build agents capable of hours-long, autonomous research work. OpenAI is developing its own Deep Research mode within its GPT model family, and Anthropic offers similar functionality in Claude. Google compares its agent's results to benchmarks such as BrowseComp, where competing models, including GPT-5.4 Pro and Opus 4.6, also post strong scores, though cross-vendor comparison methodology is rarely fully transparent.

For Polish companies and analytics teams using Google Cloud, this brings a new tool for tasks that previously required many hours of analyst work: competitive reviews, synthesis of industry reports, or preliminary due diligence on deals. The barrier to entry remains technical, though: the agent is available only via API, there's no simple consumer interface yet for the Max version, and the cost of a single complex task can exceed a typical chatbot subscription.

Google has not disclosed how many developers are currently using the preview version or exactly when broader availability on Google Cloud is planned. The company does say it plans to keep expanding integrations with industry data providers, following the model of its current partnership with FactSet, S&P Global and PitchBook.