Google is unleashing its most powerful Gemini model today, but you probably won’t be able to try it. After revealing Gemini 2.5 Deep Think at the I/O conference back in May, Google is making this AI available in the Gemini app.

Google is unleashing its most powerful Gemini model today, but you probably won’t be able to try it. After revealing Gemini 2.5 Deep Think at the I/O conference back in May, Google is making this AI available in the Gemini app.

Like some other heavyweight Gemini tools, Deep Think takes several minutes to come up with an answer. This apparently makes the AI more adept at design aesthetics, scientific reasoning, and coding. Google has exposed Deep Think to the usual battery of benchmarks, showing that it surpasses the standard Gemini 2.5 Pro and competing models like OpenAI o3 and Grok 4. Deep Think shows a particularly large gain in Humanity’s Last Exam, a collection of 2,500 complex, multi-modal questions that cover more than 100 subjects. Other models top out at 20 or 25 percent, but Gemini 2.5 Deep Think managed a score of 34.8 percent.

Mathematics is a big focus of Deep Think, which also demonstrates strong performance in the AIME benchmark. There’s still more work to be done here, though. Google recently revealed that it used a specially trained version of Deep Think, which can churn for hours before coming up with a solution, to compete in the International Mathematical Olympiad (IMO). This model earned an IMO gold medal for the first time. Google has only distributed the IMO version of Deep Think to trusted testers, but it hopes to release it more widely later. In the meantime, the standard Deep Think still reaches bronze medal status in the 2025 IMO test.

Google AI Ultra subscribers will be able to access Deep Think starting today in the Gemini app and web interface, but it doesn’t get a place in the main model menu. It’s accessible as a tool (along with Deep Research, Canvas, and others) when you select Gemini 2.5 Pro. Even with Google’s pricey AI subscription, Google says there is a set limit on the number of Deep Think queries per day. It doesn’t specify what that limit is, and Google isn’t offering specifics, suggesting that the limit will change over time. Deep Think will eventually come to the API, giving developers a way to access more prompts as a paid service.