Google has rolled out Deep Think, a powerful new gizmo for super-advanced AI users, now available through the Gemini app to Ultra subscribers. Though rooted in the same high-level framework that delivered outstanding IMO results, Deep Think takes a different path from prior Gemini 2.5 builds, streamlining that power for faster, more practical use. Deep Think employs a technique called parallel thinking, which enhances problem-solving capabilities remarkably well underneath its sleek surface. It explores multiple ideas simultaneously rather than adhering steadfastly to one narrow path in search of tentative answers. It flexes reasoning on the fly and scores big in thorny tasks like gnarly coding conundrums, scientific analysis, and multistage design brain twisters particularly well.
Google highlighted in a recent announcement that this update represents a significant stride forward from the original Gemini 2.5 showcased at I/O. The model achieves Bronze-level performance on the 2025 IMO benchmark and stays remarkably fast for most daily tasks, practically. Google CEO Sundar Pichai announced the rollout on X (formerly Twitter), saying, “We’re bringing Deep Think version that achieved gold-medal status at IMO to Ultra subscribers in @Geminiapp.”
To access Deep Think, subscribers must choose the 2.5 Pro model within the Gemini app and activate the toggle, enabling deep analysis of complex scientific and technical texts. Users get a limited number of Deep Think queries daily at launch, but Google plans to expand access gradually for select researchers and developers through API integration. Google notes that this model filters content more objectively and better than earlier versions on the safety front nowadays, apparently. It can be overly cautious sometimes and decline safe requests outright rather quietly.
