After previewing at I/O 2025 in May, Google is now rolling out the Gemini 2.5 Deep Think mode to AI Ultra subscribers.

Google says “Deep Think pushes the frontier of thinking capabilities by leveraging parallel thinking techniques.”

This approach enables Gemini to generate many ideas at once and consider them simultaneously, even revising or combining different ideas over time, before arriving at the best answer.

Gemini is given more inference or “thinking time” to “explore different hypotheses, and arrive at creative solutions to complex problems.” Google “developed novel reinforcement learning techniques that encourage the model to make use of these extended reasoning paths.”

…thus enabling Deep Think to become a better, more intuitive problem-solver over time.

Deep Think can produce “much longer responses” and leverages code execution, Google Search, and other tools.

Advertisement – scroll for more content

Today’s release offers a “significant improvement over what was first announced” in May. Google credits feedback from early trusted testers and “research breakthroughs” for these “key benchmark improvements” across coding, science, knowledge, and reasoning. Gemini 2.5 Deep Think now scores 87.6% on LiveCodeBench (competition-level coding) versus 80.4% in May.

In terms of utility, Deep Think excels at iterative development and design of “tasks that require building something complex, piece by piece.”

…we’ve observed Deep Think can improve both the aesthetics and functionality of web development tasks.

It performs well at “tough coding problems where problem formulation and careful consideration of tradeoffs and time complexity is paramount.” Google sees Deep Think as being a powerful tool for highly complex scientific and mathematical problems: “It can help formulate and explore mathematical conjectures or reason through complex scientific literature, potentially accelerating the path to discovery. “

For example, this Deep Think release can score Bronze-level performance in the 2025 International Mathematical Olympiad (IMO) benchmark and practically be used day-to-day. In comparison, the “variation” of the Deep Think model (which Google is now releasing to some academics) that was Gold-medal level took hours.

If you’re paying for Google AI Ultra ($249.99 per month in the US), Deep Think is enabled from a new button in the prompt bar when using Gemini 2.5 Pro. Subscribers will get “a few prompts” per day on mobile and web.

In the coming weeks, Google will offer a with and without tools version to trusted testers via the Gemini API to explore developer and enterprise use cases.