Google AI Ultra subscribers now have access to Deep Think, Google’s most advanced reasoning model yet for Gemini 2.5. The new model was first unveiled at Google I/O, and after feedback from "early trusted testers," is now being rolled out to a wider audience. For now, access is limited to the highest-tier AI subscription Google offers, which will run you a cool $250 per month.
The public version of Deep Think is a variation of the model that recently achieved a gold-medal standard at the International Mathematical Olympiad. Google says the version available today would achieve a bronze-medal standard at the same competition according to internal testing, and is faster and better suited for daily use. The gold-medal model will still be tested further and is being shared with a small group of mathematicians and academics for research.
Deep Think works by leveraging what Google describes as parallel thinking techniques. This approach allows Gemini to generate multiple ideas at once and consider them simultaneously. Deep Think is also multi-modal, meaning it can take in different types of data including text, images and sound.
Deep Think was tested against various AI benchmarks, such as Humanity’s Last Exam and LiveCodeBench and outscored OpenAI’s o3, xAI’s Grok 4 and Gemini 2.5 Pro, according to Google.
AI Ultra subscribers can begin using Deep Think within the Gemini app today, though they will be limited to a fixed number of prompts per day. Google did not share exactly how many prompts will be allowed.
This article originally appeared on Engadget at https://www.engadget.com/big-tech/google-rolls-out-its-most-powerful-gemini-model-yet-151205875.html?src=rss
