Google hadn’t just won a medal at the International Mathematics Olympiad, but it’s also making a version of the the model available to its users.
“We’re bringing a version of Deep Think that achieved gold-medal status at IMO to Ultra subscribers in the Gemini app,” Google CEO Sundar Pichai posted on X. He added that the official version of the model has been given to mathematicians to play around with. “Toggle it on when reasoning through complex scientific literature, tackling a coding problem that requires careful consideration of time complexities – or anything else Demis Hassabis considers a fun Friday night:)” he joked.
The Google Deep Think model predictably smashes benchmarks. It scores 34.8% on Humanity’s Last Exam, compared to 21.6% achieved by Gemini 2.5 Pro and 25.4% achieved by Grok 4.
And the model isn’t just for esoteric math problems — on LiveCodeBench, Google’s Deep Think manages 87.6%, ahead of the 79% managed by Grok 4, and 74.2% managed by Google’s Gemini 2.5 Pro.
Apart from having a near 2x improvement on the IMO benchmark, Google’s Deep Think nearly saturated the AIME 2025 benchmark with a 99.2% score. The previous best on this benchmark had been a 91.7% achieved by Grok 4.
Google had grabbed headlines last month when its official entry into the International Mathematics had delivered a gold-medal winning performance. While OpenAI had also claimed to deliver a similar result, it hadn’t officially entered the competition, and its solutions hadn’t been ratified by the Olympiad’s organizers. Google, on the other hand, had officially managed to deliver a gold medal winning performance at the event.
And it’s remarkable that a model that can achieve a gold medal at the Math Olympiad can now be accessed for $250 a month. This would’ve been unthinkable just a couple of years ago, both for the ability, and for the associated price point. And making all this intelligence available for as much a month’s worth of morning coffee could impact all manner of industries and processes in a way that’s hard to predict at this time.