Google Brings Math Olympiad Gold Winning 'Deep Think' Model to Gemini App
To access Deep Think, users must toggle the feature when selecting Gemini 2.5 Pro in the app.
Google is rolling out Deep Think, a powerful new AI feature, to Gemini app users subscribed to Google AI Ultra, bringing advanced reasoning capabilities directly to the fingertips of everyday users.
"Today, we’re making Deep Think available in the Gemini app to Google AI Ultra subscribers – the latest in a lineup of extremely capable AI tools and features made exclusively available to them," the company said.
Built on a variant of the Gemini 2.5 model that recently achieved gold at the 2025 International Mathematical Olympiad, Deep Think balances speed and usability while still reaching bronze-level IMO benchmark performance in internal testing. Google said the model has been significantly improved since its I/O announcement based on trusted tester feedback and benchmark gains.
Interestingly, during the same time, OpenAI also announced that its latest experimental reasoning model has achieved a gold-medal level performance at the 2025 International Math Olympiad (IMO)—a first in AI history.
We’re bringing a version of Deep Think that achieved gold-medal status at IMO to Ultra subscribers in the @Geminiapp (+ the official version is now in the hands of mathematicians).
— Sundar Pichai (@sundarpichai) August 1, 2025
Toggle it on when reasoning through complex scientific literature, tackling a coding problem that… pic.twitter.com/OyFSGsQSgJ
Deep Think solved five out of six problems under authentic exam conditions: two 4.5‑hour sessions, no internet access, and official problem statements, earning 35/42 points, as confirmed by three former IMO medalists.
The model excels in complex problem solving, from scientific research and mathematical conjectures to algorithm design, code generation, and iterative design tasks. Its “parallel thinking” approach enables it to explore multiple solutions simultaneously, revise them, and arrive at thoughtful answers — mimicking how humans brainstorm and refine ideas.
Deep Think also outperforms rivals in key benchmarks such as LiveCodeBench V6 and Humanity’s Last Exam, and is already being tested by mathematicians for advanced research. “We gave Deep Think in the Gemini app to mathematicians like Michel van Garrel to test mathematical conjectures,” Google shared.
To access Deep Think, users must toggle the feature when selecting Gemini 2.5 Pro in the app. Google is also expanding access through the Gemini API to select testers, while reinforcing its commitment to safety, usability, and responsible AI deployment.
