TL;DR

Google has released a significantly enhanced Gemini Deep Research agent via the Interactions API, allowing developers to embed autonomous research capabilities directly into their applications. The company has also open-sourced DeepSearchQA, a new benchmark for evaluating web research agents.

State-of-the-Art Research Performance

The Deep Research agent uses Gemini 3 Pro as its reasoning core, specifically trained to reduce hallucinations and maximise report quality during complex tasks. Google claims state-of-the-art results on multiple benchmarks: 46.4% on Humanity’s Last Exam (HLE), 66.1% on DeepSearchQA, and 59.2% on BrowseComp.

The agent operates by iteratively planning investigations - formulating queries, reading results, identifying knowledge gaps, and searching again. This release features significantly improved web search capabilities, allowing it to navigate deep into sites for specific data.

Real-World Applications Emerging

Early adopters are already seeing practical benefits. Financial firms are using Deep Research to automate initial stages of due diligence, aggregating market signals and compliance risks. In biotech, Axiom Bio reports the agent has “unlocked an unprecedented level of initial research depth” across biomedical literature, accelerating drug discovery pipelines.

Looking Forward

Deep Research will soon be available in Google Search, NotebookLM, and Google Finance. Future updates will include native chart generation for visual reports and Model Context Protocol (MCP) support for custom data sources. Google is also bringing the capability to Vertex AI for enterprise customers.

Source: Google