Posted in

Try Deep Think in the Gemini app

by Google DeepMind•

Exciting news for Google AI Ultra subscribers! Deep Think is now available in the Gemini app, offering a powerful tool for tackling complex problems. This release builds on feedback from early testers and research breakthroughs, delivering significant improvements over previous versions. Notably, it’s a variation of the model that achieved gold-medal standards at the International Mathematical Olympiad (IMO), optimized for faster, daily use while maintaining strong performance.

Deep Think Header

Deep Think leverages parallel thinking techniques, allowing Gemini to generate and evaluate multiple ideas simultaneously. By extending “thinking time,” it explores diverse hypotheses and refines solutions, supported by reinforcement learning to enhance problem-solving intuition over time.

This tool excels in areas requiring creativity and strategic planning:

  • Iterative Development and Design: It improves aesthetics and functionality in tasks like web development, delivering detailed and thoughtful responses.

Pagoda Example

  • Scientific and Mathematical Discovery: Researchers can use it to formulate conjectures or analyze complex literature, accelerating discoveries.
  • Algorithmic Development and Code: It tackles tough coding problems with careful trade-off considerations, achieving state-of-the-art results on benchmarks like LiveCodeBench V6 and Humanity’s Last Exam.

Benchmarks

Safety remains a priority, with Gemini 2.5 Deep Think showing improved content safety and objectivity, though it may sometimes refuse benign requests. Detailed safety outcomes are available in the model card for transparency.

To use Deep Think, Google AI Ultra subscribers can toggle it on in the Gemini app’s prompt bar when selecting the 2.5 Pro model. It integrates seamlessly with tools like code execution and Google Search, producing longer, more nuanced responses. Future API releases for trusted testers will further explore developer and enterprise applications.

This innovation reflects a commitment to advancing AI responsibly, empowering users to push the boundaries of knowledge and creativity.