.svg){kind=spacer}
Details
- Google has launched Gemini Deep Research for developers through the new Interactions API, allowing seamless integration of advanced autonomous research capabilities into applications. Access is provided via Google AI Studio using a Gemini API key.
- In tandem, Google has open-sourced DeepSearchQA, a benchmark featuring 900 carefully designed, multi-step research tasks spanning 17 disciplines, built to test the depth and breadth of AI agents in complex web search and information retrieval.
- The Gemini Deep Research agent leverages Google’s latest reasoning model and is tailored for sophisticated synthesis of long-context information and iterative search, autonomously generating queries, evaluating results, identifying knowledge gaps, and refining research in real time.
- Performance benchmarks show Gemini Deep Research ahead of competitors, scoring 46.4% on Humanity’s Last Exam, 66.1% on DeepSearchQA, and 59.2% on BrowseComp, surpassing OpenAI Deep Research and Perplexity Deep Research in head-to-head comparisons.
- Early use cases in finance and life sciences, such as Axiom Bio’s drug toxicity forecasting, demonstrate how Deep Research can automate critical early research steps, streamline due diligence, and accelerate scientific progress while reducing costs for enterprises.
Impact
Google’s dual release sharpens its competitive edge in the fast-evolving AI research agent space, directly challenging rivals like OpenAI and Perplexity. By granting open access to both the agent and benchmarking tools, Google aims to catalyze broader enterprise adoption and set a new standard for rigorous evaluation in knowledge-driven industries.
