Google has unveiled a major upgrade to Gemini 3 Deep Think, a specialized reasoning mode designed to tackle complex challenges in science and engineering. Developed in close collaboration with researchers, the update targets open-ended research problems where clear-cut solutions are absent and data is often incomplete or unstructured. The system combines deep scientific expertise with applied engineering tools, with a strong focus on real-world use.
The upgraded Deep Think significantly strengthens its mathematical and algorithmic capabilities. The model has set new records across several leading benchmarks, including ARC-AGI-2 and Humanity’s Last Exam, achieved a high level of performance in competitive programming, and delivered gold-medal-level results at the 2025 International Mathematical Olympiad. It also supports research-level mathematical exploration.
Beyond mathematics and programming, Gemini 3 Deep Think demonstrates strong performance in physics and chemistry, earning gold-level results in the written rounds of the 2025 international Olympiads. The system is capable of advanced analytical work, integrating abstract data with visual models. In particular, the Wang Lab at Duke University Pratt School of Engineering is using these capabilities to design next-generation semiconductor materials.
The update is now available to Google AI Ultra subscribers through the Gemini app. In addition, selected researchers, engineers, and enterprises can test the system via API as part of an early access program.