Executive Summary
Google has launched a significantly more powerful version of its Gemini Deep Research agent, making it available to developers for the first time through the new Interactions API. This agent, powered by Gemini 3 Pro, is designed to automate complex, long-running research tasks by synthesizing information from both the web and user-uploaded documents into detailed, citable reports. Alongside this release, Google has open-sourced DeepSearchQA, a new benchmark designed to test the comprehensiveness of agents on complex web research tasks.
Key Takeaways
* Product: Gemini Deep Research, an AI agent optimized for autonomous, multi-step research and report generation.
* Access: Available now to developers via the new Interactions API, accessible with a Gemini API key from Google AI Studio.
* Core Technology: Built on Gemini 3 Pro and trained with multi-step reinforcement learning to improve accuracy, reduce hallucinations, and autonomously navigate complex information.
* Key Capabilities:
* Synthesizes information from public web data and user-uploaded files (PDFs, CSVs, docs).
* Provides detailed, granular citations for generated claims.
* Supports structured outputs, including JSON schema, for easy integration into applications.
* Allows developers to control the output structure and formatting via prompting.
* New Benchmark: Google has open-sourced DeepSearchQA, a new benchmark with 900 tasks to evaluate an agent's ability to perform intricate, multi-step research.
* Future Availability: The agent will be integrated into Google Search, NotebookLM, Google Finance, and the Gemini App, and will be brought to Vertex AI for enterprise customers.
Strategic Importance
This release productizes Google's advanced agentic AI research, aiming to drive developer adoption of the Gemini platform by enabling the creation of powerful, automated research tools for industries like finance and biotech.