Gemini 3 Deep Think has been significantly upgraded to tackle complex challenges in science, research, and engineering. The enhanced reasoning mode is now available to Google AI Ultra subscribers in the Gemini app and, for the first time, is accessible via the Gemini API through an early access program for researchers, engineers, and enterprises. Developed in close collaboration with scientists and researchers, the new Deep Think is designed to handle real-world problems where clear solutions are rare, data is incomplete, and uncertainty is high. It combines deep scientific knowledge with practical engineering applications, moving beyond theoretical reasoning to deliver tangible results. Early adopters are already seeing powerful outcomes. Lisa Carbone, a mathematician at Rutgers University, used Deep Think to analyze a highly technical paper in theoretical physics. The model identified a subtle logical flaw that had gone unnoticed in human peer review. At Duke University, the Wang Lab leveraged Deep Think to optimize crystal growth techniques for semiconductor materials. The model successfully generated a recipe to produce thin films over 100 micrometers in size—meeting a challenging target that previous methods struggled to achieve. Anupam Pathak, an R&D lead at Google’s Platforms and Devices division, used Deep Think to accelerate the design of physical components, demonstrating its ability to apply rigorous mathematical and algorithmic reasoning. The latest version of Deep Think has achieved top-tier performance across some of the most demanding benchmarks. It scored 48.4% on Humanity’s Last Exam—without using external tools—setting a new standard. It reached 84.6% on ARC-AGI-2, verified by the ARC Prize Foundation, and achieved an Elo rating of 3455 on Codeforces, a leading competitive programming platform. It also delivered gold-medal-level results on the 2025 International Math Olympiad. In scientific domains, Deep Think excels in physics and chemistry. It performed at a gold-medal level on the written sections of the 2025 International Physics and Chemistry Olympiads and scored 50.5% on the CMT-Benchmark, a test of advanced theoretical physics knowledge. Beyond academic excellence, the model is built for real-world impact. It can interpret complex data, simulate physical systems through code, and even convert hand-drawn sketches into 3D-printable models. By analyzing the sketch, generating a precise 3D model, and producing a ready-to-print file, Deep Think bridges the gap between idea and physical reality. Google is now opening access to this advanced reasoning mode through the Gemini API, with early access available to qualified researchers and organizations. The goal is to empower innovators at the frontiers of science and engineering to push discovery further. Google AI Ultra subscribers can start using the updated Deep Think today in the Gemini app. The company invites the research and engineering communities to join the early access program and help shape the future of intelligent systems.