Google has launched Deep Think, its most advanced Gemini reasoning model, exclusively for AI Ultra subscribers at $250 monthly. This rollout follows initial testing with select users.
Deep Think debuted at Google I/O earlier this year. After incorporating feedback from early trusted testers, the model is now accessible to all AI Ultra tier subscribers. Google’s $250 monthly subscription represents its highest-priced AI access tier.
The public release constitutes a modified version of a model that attained gold-medal level performance at the International Mathematical Olympiad. Internal evaluations indicate this iteration achieves a bronze-medal standard at the same competition. Google optimized this variant for daily usability and processing speed. The original gold-medal model remains under evaluation with mathematicians and academics for research purposes.
Google employs parallel thinking techniques within Deep Think, enabling simultaneous generation and consideration of multiple ideas. The model processes multimodal inputs, including text, images, and audio.
In benchmark assessments—specifically Humanity’s Last Exam and LiveCodeBench—Deep Think surpassed scores from OpenAI’s o3, xAI’s Grok 4, and Google’s own Gemini 2.5 Pro, according to company testing data.
AI Ultra members can immediately utilize Deep Think via the Gemini application. Google confirmed a daily cap on prompt interactions exists but did not disclose the specific numerical limit.