Google is now amping up the AI competition against its rivals – and its response to OpenAI’s release of GPT 5.2 is the release of a revamped version of Gemini Deep Research built on its newest flagship model, Gemini 3 Pro, marking one of the company’s most aggressive steps yet into agentic AI — systems that plan, search and reason through tasks with minimal human direction.
“Gemini Deep Research is an agent optimized for long-running context gathering and synthesis tasks. The agent’s reasoning core uses Gemini 3 Pro, our most factual model yet, and is specifically trained to reduce hallucinations and maximize report quality during complex tasks. By scaling multi-step reinforcement learning for search, the agent autonomously navigates complex information landscapes with high accuracy. Deep Research iteratively plans its investigation – it formulates queries, reads results, identifies knowledge gaps, and searches again. This release features vastly improved web search, allowing it to navigate deep into sites for specific data,” Google announced in an official blog post.
For those who missed it, the original version of Gemini Deep Research, launched last year atop Gemini 1.5 Pro, functioned largely as a customizable report-writing assistant. The new release transforms it into a full autonomous researcher that can ingest vast data sets, retrieving information from the open web, and producing structured analyses that unfold across multiple reasoning steps. This means that rather than combing through results, individuals may increasingly rely on AI systems to identify relevant material, verify details and synthesize findings into structured reports. This could substantially reduce the time that students, researchers and professionals spend aggregating information, especially in areas where accuracy and source attribution matter.
Google says Gemini 3 Pro’s improved visual and symbolic reasoning allows the agent to extract facts from handwriting, equations, complex charts, and other formats that often trip up AI models. The company claims the agent can now navigate deep within websites, correlate findings across documents, and refine its searches iteratively — planning an investigation rather than simply responding to a prompt. Enterprise users stand to see an even more pronounced change. Long-running due-diligence reviews, regulatory compliance checks and technical investigations could be delegated to autonomous workflows that operate across internal databases and external sources with minimal human prompting. The Interactions API’s control features — including source whitelists, step-level transparency and structured citations — offer a level of governance that aligns with the expectations of legal, financial and scientific teams.
The launch coincides with Google’s debut of the Interactions API, a new access point designed to unify how developers call Gemini models and Google-built agents. The API standardizes long-running workflows, simplifies data management, and enables developers to create their own agents that can chain tools, browse, and reason over extended periods. Google says future built-in agents will be added to the API, alongside developer tools for connecting AI systems to external databases or third-party applications.
To validate the upgrade, Google tested Gemini Deep Research on several benchmarks, including HLE (Humanity’s Last Exam, which is a 2,500-question general knowledge and technical reasoning test where more than half the items involve math, physics, or programming. Google reports a 46.4% solve rate), DeepSearchQA (a newly open-sourced, multi-step research benchmark with 900 tasks requiring sequential reasoning), and BrowserComp (a browser-based fact-finding test used to evaluate agents navigating real webpages). Google says the new agent achieved state-of-the-art results on HLE and DeepSearchQA and surpassed earlier Google models on BrowserComp. However, OpenAI’s now-released GPT-5.2 edged ahead of Google on BrowserComp and, according to OpenAI, surpasses rivals broadly on its own evaluation suite.
Google says firms are already using Deep Research for due diligence, scientific safety analysis, competitive intelligence, and other workloads requiring deep context and slow reasoning. The company confirmed it will embed the agent into several consumer products — including Google Search, Google Finance, the Gemini app, and NotebookLM — signaling a future where AI agents, rather than users, perform complex web queries. Google’s timing was deliberate. With the AI community anticipating OpenAI’s GPT-5.2, Google positioned its upgraded agent as evidence that it remains competitive in agentic research tech. In addition to this, developers gain new governance controls through the Interactions API, including constraints on data sources, required citations, and modular reasoning paths — all aligned with enterprise and regulatory frameworks for high-risk AI systems.