Deep Think is now available in the Gemini app for Google AI Ultra subscribers. This tool is the latest addition to a suite of powerful AI features offered exclusively to these subscribers.
This new version integrates insights from early testers and recent research advancements. It represents a notable enhancement compared to what was initially unveiled at I/O, showing improvements in key benchmarks and positive feedback from testers. This model is a variant of the one that recently earned a gold-medal standard at this year’s International Mathematical Olympiad (IMO). While the IMO-level model requires hours for complex mathematical reasoning, the current release is designed for faster, daily use, achieving Bronze-level performance on the 2025 IMO benchmark, according to internal assessments.
Deep Think could be a powerful tool in creative problem solving:
In addition to making Deep Think available to Google AI Ultra subscribers, the official Gemini 2.5 Deep Think model, which achieved the gold-medal standard, is being shared with a select group of mathematicians and academics. Their insights on how it can enhance research and inquiry are anticipated, and their feedback will be utilized for ongoing improvements.
This release marks a significant advancement in developing more helpful and capable AI, reinforcing the commitment to using Gemini to expand the boundaries of human knowledge.
How Deep Think works: extending Gemini’s parallel “thinking time”
Similar to how individuals approach complex problems by exploring various angles, evaluating solutions, and refining their answers, Deep Think advances thinking capabilities through parallel thinking techniques. This method allows Gemini to generate and evaluate multiple ideas concurrently, even revising or merging them over time, to arrive at the optimal solution.
Furthermore, by extending its inference or “thinking time,” Gemini is afforded more opportunity to explore various hypotheses and develop creative solutions for complex problems.
Novel reinforcement learning techniques have also been developed to encourage the model to utilize these extended reasoning paths, thereby enabling Deep Think to evolve into a more effective and intuitive problem-solver.
How Deep Think stacks up: state-of-the-art performance
Deep Think can help people tackle problems that require creativity, strategic planning and making improvements step-by-step, such as:
- Iterative development and design: Deep Think demonstrates impressive performance on tasks involving complex, piece-by-piece construction. For instance, it has been observed to enhance both the aesthetics and functionality of web development projects.
Deep Think in the Gemini app uses parallel thinking techniques to deliver more detailed, creative and thoughtful responses.
- Scientific and mathematical discovery: Its ability to reason through highly complex problems makes Deep Think a valuable asset for researchers. It can assist in formulating and exploring mathematical conjectures or analyzing intricate scientific literature, potentially speeding up the discovery process.
- Algorithmic development and code: Deep Think shows particular strength in challenging coding problems where precise problem formulation and careful evaluation of tradeoffs and time complexity are crucial.
Deep Think’s performance is also reflected in challenging benchmarks that measure coding, science, knowledge and reasoning capabilities. For example, compared to other models without tool use, Gemini 2.5 Deep Think achieves state-of-the-art performance across LiveCodeBench V6, which measures competitive code performance, and Humanity’s Last Exam, a challenging benchmark that measures expertise in different domains, including science and math.
Advancing Gemini Responsibly
Safety and responsibility are continuously integrated into Gemini throughout its training and deployment. During testing, Gemini 2.5 Deep Think exhibited enhanced content safety and tone-objectivity compared to Gemini 2.5 Pro, though it showed a greater inclination to decline benign requests.
As Gemini’s problem-solving capabilities progress, a thorough examination of risks associated with increased complexity is underway. This includes frontier safety evaluations and the implementation of planned mitigations for critical capability levels.
Further details on the safety outcomes of Gemini 2.5 Deep Think are available in the model card.
How to use Deep Think in the Gemini app today
If you’re a Google AI Ultra subscriber, you can use Deep Think in the Gemini app today with a fixed set of prompts a day by toggling “Deep Think” in the prompt bar when selecting 2.5 Pro in the model drop down. Deep Think automatically works with tools such as code execution and Google Search, and can produce much longer responses.
Efforts are also underway to release Deep Think, both with and without tools, to a group of trusted testers via the Gemini API in the coming weeks. This aims to better understand its usability for developer and enterprise applications.
Teams at nearly every layer of the stack, from research to deployment, have worked to make Deep Think faster, more reliable, and user friendly for Gemini app users.
