Google AI Extremely subscribers now have entry to Deep Assume, Google’s most superior reasoning mannequin but for Gemini 2.5. The brand new mannequin was first unveiled at Google I/O, and after suggestions from “early trusted testers,” is now being rolled out to a wider viewers. For now, entry is proscribed to the highest-tier AI subscription Google affords, which is able to run you a cool $250 monthly.
The general public model of Deep Assume is a variation of the mannequin that lately achieved a gold-medal normal on the Worldwide Mathematical Olympiad. Google says the model out there at this time would obtain a bronze-medal normal on the identical competitors in line with inner testing, and is quicker and higher fitted to each day use. The gold-medal mannequin will nonetheless be examined additional and is being shared with a small group of mathematicians and lecturers for analysis.
Deep Assume works by leveraging what Google describes as parallel pondering methods. This strategy permits Gemini to generate a number of concepts directly and think about them concurrently. Deep Assume can be multi-modal, that means it will probably soak up various kinds of knowledge together with textual content, photographs and sound.
Deep Assume was examined in opposition to numerous AI benchmarks, reminiscent of Humanity’s Final Examination and LiveCodeBench and outscored OpenAI’s o3, xAI’s Grok 4 and Gemini 2.5 Professional, in line with Google.
AI Extremely subscribers can start utilizing Deep Assume inside the Gemini app at this time, although they are going to be restricted to a set variety of prompts per day. Google didn’t share precisely what number of prompts will likely be allowed.
