Google has open sourced DeepSearchQA and opened access to its Gemini Deep Research Agent.
Google has opened access to its Gemini Deep Research Agent for the first time, allowing developers to integrate advanced autonomous research capabilities directly into their applications. Powering the upgraded agent is the Gemini 3 Pro model, which Google says delivers a significant leap in reasoning depth, autonomy, and reliability.
Designed to operate like a human researcher, the Deep Research Agent performs multi-step web searches, refines follow-up questions, validates sources, identifies information gaps, and continues searching until a complete conclusion is reached. Google states this approach produces deeper and more reliable insights than traditional single-prompt, single-response AI systems.
A key announcement accompanying the launch is DeepSearchQA, an open-source benchmark created to evaluate how well AI systems handle complex, multi-step reasoning. The benchmark includes 900 tasks across 17 subjects, including climate, history, and health. Each task builds on the previous one, enabling assessment of an AI system’s ability to maintain logic, consistency, and accuracy across extended reasoning chains. Google positions DeepSearchQA as a shared, open standard to push industry-wide progress in advanced AI reasoning.
Developers also gain access to new capabilities through the Deep Research API, including document parsing, structured report templates, and detailed source citations, designed for easy integration into existing workflows. Google confirmed upcoming enhancements such as native chart creation and broader support for the Model Context Protocol.
Additionally, Google launched the Interactions API, enabling dynamic agent interactions, server-side session handling, and background execution, reinforcing its strategic shift towards fully autonomous AI agents that investigate, reason, and decide rather than merely generate answers.
