This is Google’s new deep research agent, powered by Gemini 3 Pro

Google has launched the Gemini Deep Research agent, which is designed to work autonomously on the web and leverage the most advanced deep research capabilities of the Gemini 3 Pro artificial intelligence (AI) model.

The new agent is based on Google’s most advanced model, Gemini 3 Pro, allowing it to “autonomously plan, execute and synthesize multi-step research tasks,” as the company explains on the developers page.

With these capabilities, you can search and use information from web environments in which you navigate autonomously and from the user’s own data to conduct research, which you present in reports with detailed citations.

Precisely, its main characteristic is deep research, which it carries out iteratively, as it formulates queries, reads results, identifies knowledge gaps and carries out new searches.

Google has highlighted in a statement the power of Gemini’s Deep Research agent, which is supported by the results it has obtained in benchmarks such as Humanity’s Last Exam, where it obtains 46.4%; in DeepSearchQA, with 66.1%; and in BrowseComp, in which it reaches 59.2%.

In these three evaluations, the capabilities of GPT-5 Pro have also been tested, which obtains 38.9%, 65.2% and 59.5% percent, respectively; and Gemini 3 Pro, which achieves 43.2%, 56.6% and 49.4%, respectively.

The Gemini Deep Research agent is available to developers with the new Interactions API, available in the Gemini API from Google AI Studio. According to Google, it will also soon arrive in Google Search, NotebookLM, Google Finance and will be updated in the Gemini ‘app’.