Gemini, the artificial intelligence platform developed by Google DeepMind, has introduced Gemini Deep Research and DeepSearchQA, its two newest and most powerful optimized AI agents designed to perform complex information search and analysis tasks, minimizing hallucinations.
Gemini Deep Research is engineered to optimize the collection and synthesis of information, enabling developers to integrate autonomous research into their applications through a new Interactions API. For its part, DeepSearchQA establishes a new standard for evaluating agents in web search tasks that require multiple steps, improving the accuracy and thoroughness of the results generated.
Both of these agents operate with the Gemini 3 Pro model and are capable of handling extensive contexts, identifying knowledge gaps, and conducting complex searches autonomously.
Gemini Deep Research has been enhanced to carry out more complex research tasks that require in-depth analysis and the collection of data from diverse sources. It conducts web searches with greater accuracy and generates comprehensive reports. This agent plans its research iteratively: it formulates queries, reads results, identifies gaps in knowledge, and conducts new searches as required.
Additionally, thanks to its integration with the Gemini 3 Pro model (which now leads nearly every LMArena category), it has achieved improved synthesis capabilities and further reduced hallucinations.
Its most significant feature is the ability to handle long and complex contexts, which allows developers to integrate extensive information into the responses generated. With the new “Interactions” API, developers may incorporate these capabilities into their own applications, facilitating the creation of advanced research tools.
“Deep Research is now more useful and intelligent than ever, and it will soon be available in Google Search, NotebookLM, Google Finance, and will be updated in the Gemini app,” the company stated.
DeepSearchQA is a new benchmark that sets a reference point for assessing the accuracy of agents in web search tasks requiring multiple reasoning steps. Unlike traditional tests that measure factual correctness, DeepSearchQA measures thoroughness, enabling agents to generate more comprehensive and detailed responses.
This benchmark includes 900 “causal chain” tasks designed across 17 fields, where each step depends on analysis performed in the previous one. In contrast to standard fact-based evaluations, DeepSearchQA measures thoroughness, which requires agents to produce detailed answer sets. This evaluates both research accuracy and data retrieval capabilities.
This agent also functions as a diagnostic tool to assess the benefits of reflection time. “In our internal evaluations, we observed significant performance improvements when allowing the agent to perform additional searches and reasoning steps, which we plan to explore in future versions,” Google explained.
The Gemini Deep Research and DeepSearchQA agents are already delivering remarkable results in several sectors, including financial services, biotechnology, and market research. In the financial sector, these agents are used to automate the initial stages of due diligence, reducing the time required for research and improving decision-making processes.
In the field of biotechnology, Gemini Deep Research has proven to be valuable in biomedical research, accelerating drug discovery processes by collecting and analyzing data with greater accuracy than traditional methods.
“Gemini Deep Research generates granular data and evidence far beyond what a human researcher alone could previously achieve. We are excited to build upon this as a foundation for agentic systems that reason from molecular mechanisms to experimental data and clinical outcomes, empowering scientists to develop safer medicines,” stated Alex Beatson, Co-founder of Axiom Bio, a company focused on developing AI systems for predicting drug toxicity.
Gemini Deep Research offers developers a suite of advanced tools for integrating research capabilities into their applications. Prominent functionalities include the ability to analyze documents in a variety of formats (including PDF, CSV, or DOCS), manage extensive contexts, and generate detailed reports with precise citations.
In addition, research results can be structured in formats such as JSON, making subsequent analysis within other applications easier. Future updates will include the ability to generate charts natively for visual reports, and enhanced connectivity through the Model Context Protocol (MCP), which will allow developers to access customized data sources more easily.
Your email address will not be published. Required fields are marked *
Δ