Google is now offering Deep Think in the Gemini app to Google AI Ultra subscribers, bringing the company's most advanced reasoning model to mobile users. The tool uses parallel thinking techniques—essentially generating multiple ideas simultaneously and evaluating them before settling on the best answer. Users can enable it by toggling Deep Think in the prompt bar when selecting Gemini 2.5 Pro.

This release represents a streamlined version of the model that recently achieved gold-medal standard at the International Mathematical Olympiad. While that iteration requires hours to solve complex math problems, today's version is faster and more usable for everyday tasks while maintaining Bronze-level performance on the 2025 IMO benchmark based on internal evaluations. The full gold-medal model is being shared with select mathematicians and academics for research purposes.

Deep Think particularly excels at tough coding problems requiring careful consideration of tradeoffs, iterative development and design, and scientific or mathematical discovery. Google reports state-of-the-art performance on LiveCodeBench V6, which measures competitive code execution, and Humanity's Last Exam, a challenging benchmark spanning multiple domains. The company is also working to release Deep Think via the Gemini API for developer and enterprise use cases.