Google has launched Deep Think for its Gemini app, exclusively for Google AI Ultra subscribers. This advanced AI feature enhances problem-solving capabilities by allowing deeper, longer reasoning before generating responses. Following extensive feedback from testers, Deep Think evolves from a slower, award-winning model that excelled at the 2025 International Mathematical Olympiad (IMO), to a faster, everyday tool. It performs at a bronze-medal level on IMO-style challenges, making it ideal for complex fields such as science, mathematics, coding, and design. By offering the model more “thinking time,” Deep Think can evaluate multiple ideas and refine answers through reinforcement learning techniques. Performance benchmarks, like LiveCodeBench V6 and Humanity’s Last Exam, demonstrate its capabilities in code quality and academic reasoning. Additionally, Deep Think prioritizes content safety and neutrality. Users can activate this feature in the Gemini app and benefit from extended response generation, with future testing planned for developers and enterprise users via the Gemini API.
Source link
