Google has made Deep Think available in the Gemini app for Google AI Ultra subscribers. This is the latest in the lineup of Google’s advanced AI tools and features. The search giant is also giving select mathematicians access to the full version of the Gemini 2.5 Deep Think model that entered into the International Mathematical Olympiad (IMO) competition.
ARTICLE CONTINUES BELOW VIDEO
According to the company, the latest release incorporates feedback from early trusted testers and research breakthroughs. The new model is a significant improvement over what was showcased at the Google I/O earlier this year. The model is a variation of the one that recently achieved the gold-medal standard at this year’s IMO. While the earlier model took hours to reason about complex math problems, Google claims that the latest model is faster and more usable for day-to-day activities. Based on internal evaluations, the new version reaches the Bronze-level performance on the 2025 IMO benchmark.
Google claims that Deep Think could be a powerful tool in creative problem solving. The tech giant gave access to Deep Think on the Gemini app to mathematicians like Michel van Garrel to test mathematical conjectures. Along with Deep Think for Google AI Ultra subscribers, the company is also giving access to the official version of the IMO gold-winning Gemini 2.5 Deep Think model with a small group of mathematicians and academics.
How does Deep Think work?
Much like humans who explore different perspectives to a complex problem, weigh in on likely solutions, and revise till the final answer, Deep Think deploys parallel thinking techniques. Google said that this approach allows Gemini to come up with multiple ideas at once and consider them at the same time, sometimes even revising or combining these over time, before offering the best answer.
With Deep Think, Google said it has extended the inference time, or thinking time, endowing the model with the time to explore different possibilities and come up with creative solutions to complex problems. Besides, the company has also developed novel reinforcement learning techniques that push the model to use extended reasoning paths, allowing it to become a better problem-solver over time.
When it comes to use cases, according to Google, Deep Think can help users tackle problems that need creative thinking, strategic planning, and making improvements in a step-by-step manner. When it comes to design development, the model can improve both the aesthetics and functionality of web development tasks. It can be useful in scientific and mathematical discovery, as this model can assist in formulating and exploring mathematical conjectures and can even reason through complex scientific literature. Since Deep Think is believed to excel at tough coding problems, it can be of help in algorithmic development and code.
© IE Online Media Services Pvt Ltd
