Google has rolled out a new feature named Deep Think for subscribers of its Gemini Ultra plan, aimed at boosting the model’s problem-solving capabilities through longer and more complex reasoning processes.

Now available in the Gemini app, the tool builds on a refined version of the AI model that recently secured gold-medal performance at the International Mathematical Olympiad (IMO), the company said in an official blog post.

Deep Think employs parallel reasoning techniques, allowing it to evaluate multiple possible solutions before delivering an answer. While the competition-grade version takes hours to solve complex problems, the consumer-facing tool runs more quickly and maintains bronze-level IMO performance, based on internal testing.

Google said the feature is particularly suited to tasks that demand strategic planning, such as programming, scientific analysis, and iterative design. A limited number of mathematicians have also received access to the full competition-grade model for academic research.

According to Google, safety tests show that Deep Think has better content moderation than Gemini 2.5 Pro, though it may block more benign queries. Access is currently limited to a set number of daily prompts for Ultra subscribers, with plans to offer API access to developers in the near future.

The move continues Google’s trend of releasing advanced AI features exclusively to its paid Ultra tier, though no pricing changes have been announced alongside this update.