Gemini 2.5 Deep Think

Gemini 2.5 Deep Think is a new model for Google AI Ultra subscribers. It's a version of the model that achieved a gold-medal standard at the IMO, designed to solve complex problems with parallel "thinking time".

Gemini 2.5 Deep Think is an advanced AI model for Google AI Ultra subscribers, designed to tackle complex problems with enhanced reasoning. Inspired by the gold-medal-winning model from the International Mathematical Olympiad, it employs parallel 'thinking time' to explore multiple solutions simultaneously, refining answers through extended reasoning. This version is optimized for daily use, balancing speed with bronze-level IMO performance. The model excels in creative problem-solving, iterative design, scientific research, and coding challenges. It outperforms competitors in benchmarks like LiveCodeBench V6 and Humanity’s Last Exam, offering state-of-the-art performance in coding, science, and reasoning. Safety improvements include better content moderation, though it may sometimes reject benign requests. Available in the Gemini app, Deep Think supports tools like code execution and Google Search, delivering detailed, thoughtful responses. Future releases aim to expand its usability for developers and enterprises via the Gemini API. This innovation represents a leap forward in AI-assisted problem-solving and knowledge exploration.