Google launches Deep Research in the Gemini app

Google added Deep Research to the Gemini app for Gemini Advanced subscribers, a feature that used the model’s reasoning and long-context capacity to plan a multi-step research process, browse the web to gather information, and compile the results into a written report rather than answering a query with a single response. It launched the same day as the broader Gemini 2.0 announcement, alongside prototype agents including Project Mariner and the coding agent Jules.

Where a standard chatbot response drew on a single pass of retrieval or none at all, Deep Research was built to decompose an open-ended question into a research plan, issue a series of searches, read and synthesise multiple sources, and produce a structured writeup — a workflow more like an analyst’s task than a single query-and-answer exchange. Google positioned it as a step toward agents that could carry out extended, multi-stage work with limited supervision rather than respond only to single prompts.

Deep Research arrived at a moment when several labs were converging on the same idea of an agentic research assistant that browsed autonomously and returned a synthesised report; competing versions from OpenAI and others followed over the subsequent months. Google described the December 2024 version as an early release, and continued to upgrade the feature’s underlying model and capabilities through 2025 rather than treating it as finished at launch — consistent with the pattern across Google’s December 2024 agent announcements, most of which shipped as previews to be iterated on rather than complete products.