Gemini 3 Deep Think: Advancing science, research and engineering (blog.google)

Google has launched a significant upgrade to its Gemini 3 Deep Think AI, aiming to enhance capabilities in science, research, and engineering. This specialized reasoning mode, developed in collaboration with researchers, is designed to solve complex, ambiguous problems where data is often inconsistent or incomplete. Notable early applications include a mathematician identifying a critical flaw in a physics paper and researchers optimizing semiconductor fabrication methods. Deep Think is now accessible to Google AI Ultra subscribers and select enterprises through the Gemini API, opening new avenues for academic and industrial research. The update marks a leap in AI reasoning, achieving impressive scores on rigorous benchmarks, such as 48.4% on Humanity’s Last Exam and gold-medal level performances in competitive math and science Olympiads. These advancements not only bolster rigorous academic standards but also have practical implications for engineers, enabling tasks like transforming sketches into 3D-printable files. The incorporation of Deep Think into various scientific domains, along with its capability to interpret complex data, positions it as a pivotal tool in modern engineering and research, expanding the frontier of artificial intelligence in tackling real-world challenges.