Google has unveiled a significant upgrade to Gemini’s specialized reasoning feature, Deep Think, aimed at addressing complex challenges in science, research, and engineering. This update, created in collaboration with scientists and researchers, enables Deep Think to handle intricate data scenarios often lacking clear solutions. Notable achievements of the Gemini 3 Deep Think upgrade include a record 48.4% on Humanity’s Last Exam and an exceptional 84.6% on ARC-AGI-2, according to the ARC Prize Foundation. In competitive coding, it achieved an Elo of 3455 on Codeforces and is recognized for gold-medal level performance in the International Math Olympiad 2025. The upgraded Deep Think enhances practical applications, allowing researchers to interpret complex data and enabling engineers to model systems through code. Users can transform sketches into 3D-printed objects, making this technology accessible through the Gemini app for Google AI Ultra subscribers and via the Gemini API for enterprises.
Source link
