Google has released a significantly more powerful Gemini Deep Research agent via the Interactions API, allowing developers to embed autonomous research capabilities into their applications. Alongside this release, Google is open-sourcing DeepSearchQA, a new benchmark designed to test agent comprehensiveness on complex web research tasks.
The agent utilizes Gemini 3 Pro and achieves state-of-the-art results: 46.4% on Humanity’s Last Exam (HLE), 66.1% on DeepSearchQA, and 59.2% on BrowseComp. It iteratively plans investigations by formulating queries, reading results, and identifying knowledge gaps to navigate deep into sites for specific data.
DeepSearchQA features 900 hand-crafted "causal chain" tasks across 17 fields, measuring both research precision and retrieval recall through exhaustive answer sets. Early feedback indicates the agent is being used in financial services for due diligence automation and by biotech firms like Axiom Bio to accelerate drug discovery pipelines.
Developers can access the agent via the Interactions API using a Gemini API key from Google AI Studio, with future updates planned for Vertex AI and native chart generation.