Google Unveils Gemini 3 Deep Think: AI for Science

In a groundbreaking step towards enhancing scientific and engineering applications, Google has introduced Gemini 3 Deep Think, a highly advanced AI model update. Launched on February 4, 2026, this innovative tool is poised to transform how researchers analyze complex data and how engineers model physical systems, including the ability to convert sketches into 3D-printable models. With state-of-the-art performance metrics, Gemini 3 Deep Think aims to bridge the gap between theoretical knowledge and practical application in science and engineering fields.

State-of-the-Art Performance Metrics

Gemini 3 Deep Think has set new benchmarks in various academic fields, showcasing its prowess in rigorous assessments. The model achieved a remarkable 48.4% score on Humanity’s Last Exam, a test designed to evaluate the limits of modern AI models. Furthermore, it earned an impressive 84.6% on the ARC-AGI-2, a feat verified by the ARC Prize Foundation. In competitive programming, the model reached an Elo rating of 3455 on Codeforces, demonstrating its capabilities in real-time problem-solving. Additionally, it performed at a gold-medal level in the 2025 International Math Olympiad, solidifying its status as a frontrunner in AI advancements.

Advancements in Scientific Domains

Gemini 3 Deep Think extends its capabilities beyond mathematics, excelling in diverse scientific fields such as chemistry and physics. The model showcases gold medal-level results in the written sections of the 2025 International Physics Olympiad and Chemistry Olympiad. It also demonstrates proficiency in advanced theoretical physics, achieving a score of 50.5% on the CMT-Benchmark. These accomplishments reflect the model's versatility and potential to assist researchers in navigating complex scientific domains, thereby accelerating discovery and innovation.

Access and Availability

Access to Gemini 3 Deep Think is streamlined for users through Google AI Ultra subscriptions via the Gemini app. Additionally, an Early Access Program is available for scientists, engineers, and enterprises, enabling them to utilize the Gemini API for their specific needs. This initiative not only underscores Google’s commitment to advancing AI technology but also positions Gemini 3 Deep Think as an accessible tool for professionals looking to leverage AI in real-world applications.

TLDR

The launch of Gemini 3 Deep Think marks a significant milestone in the integration of AI within scientific and engineering practices. With its record-setting performance metrics and broad applicability across various disciplines, this model is set to empower researchers and engineers alike. As Google continues to innovate and expand access to advanced AI tools, the potential for transforming theoretical research into practical solutions is more promising than ever.