Google has released a major upgrade to Gemini 3 Deep Think, its specialized reasoning mode designed to tackle complex challenges in science, research, and engineering. The enhanced AI system is now available to Google AI Ultra subscribers through the Gemini app, with select researchers and enterprises gaining early access via the Gemini API.

The update represents Google's push to move beyond abstract theory toward practical applications. Deep Think was developed in close partnership with scientists to handle messy, incomplete data and problems without clear solutions. The system has already demonstrated impressive capabilities, including identifying a subtle logical flaw in a mathematics paper that had passed human peer review, and designing crystal growth recipes for semiconductor materials that previous methods couldn't achieve.

Deep Think achieves gold-medal level performance across rigorous benchmarks, scoring 48.4% on Humanity's Last Exam and 84.6% on ARC-AGI-2. It also excels in competitive programming with an Elo rating of 3455 on Codeforces. The system can now interpret complex data, model physical systems through code, and even transform sketches into 3D-printable files. Researchers and engineers can express interest in the early access program to test these capabilities for their own projects.