Google has unveiled its most advanced reasoning model, Deep Think, for the Gemini platform, exclusively available to Google AI Ultra subscribers for $250 per month.
The Deep Think model was first introduced at Google I/O 2025 and has been rolled out widely after receiving feedback from early trusted testers. According to the company, the public variant of Deep Think is adapted from a version that achieved a gold-medal performance standard at the International Mathematical Olympiad.
In internal testing, the release version of Deep Think has been found to deliver bronze-medal capabilities while being optimized for daily usage speed and efficiency. The original gold-level model is still being evaluated with the help of mathematicians and academics.
Deep Think employs parallel thinking techniques, enabling it to process multiple ideas simultaneously. It also supports multimodal inputs, including text, images, and audio. In benchmark tests such as Humanity’s Last Exam and LiveCodeBench, Deep Think outperformed competitors like OpenAI’s o3, xAI’s Grok 4, and Google’s own Gemini 2.5 Pro.
To access Deep Think, users need a premium AI Ultra subscription. They can interact with the model via the Gemini app, although Google has imposed undisclosed daily prompt limits. The company is positioning this release as a significant advancement in AI reasoning capabilities, specifically tailored for specialized users.
