On Thursday, Google introduced a revamped version of its Gemini Deep Research agent, now powered by the company’s flagship Gemini 3 Pro model — an announcement that coincidentally landed the same day OpenAI released GPT-5.2.
The updated agent is no longer limited to producing research reports. Developers can now integrate its high-level research capabilities directly into their applications through Google’s new Interactions API, designed to give them more granular control as agentic AI becomes the norm.
Gemini Deep Research is built to digest massive volumes of information and process exceptionally large prompts. Google says customers already rely on it for tasks ranging from due diligence to drug toxicity assessments.
The company plans to bring the new agent into a range of Google products, including Search, Google Finance, the Gemini app, and NotebookLM — a step toward a future where users may rely on AI agents instead of traditional search.
According to Google, Deep Research benefits heavily from Gemini 3 Pro, which the company calls its “most factual” model and one specifically engineered to reduce hallucinations in complex workflows. Hallucination risk is particularly high in long, multi-step agentic processes, where a single incorrect decision can derail the entire output.
To demonstrate progress, Google introduced a new benchmark called DeepSearchQA, designed to test agents on sophisticated, multi-step information-seeking tasks. The benchmark has been open-sourced.
Google also evaluated Deep Research on Humanity’s Last Exam, a notoriously difficult general-knowledge benchmark, and BrowserComp, which measures browser-based agent skills.
Unsurprisingly, Google’s agent led the field in Google’s own tests and performed best on Humanity’s Last Exam. However, OpenAI’s ChatGPT 5 Pro came close across the board and edged ahead on BrowserComp.
Those comparisons were outdated almost immediately, as OpenAI launched its long-awaited GPT-5.2 — codenamed “Garlic” — on the very same day. OpenAI claims the model outperforms competitors, including Google, on a wide range of standard benchmarks.
The timing of Google’s announcement raised eyebrows, as the company unveiled its new research agent just as the industry’s attention was fixed on OpenAI’s next major release.
