Google has released an updated version of Gemini 3 Deep Think, a specialized reasoning mode designed to address complex problems in science, research, and engineering.
Gemini 3 Deep Think
The upgrade was developed in collaboration with scientists and researchers to handle challenges with incomplete data, unclear boundaries, and no single correct solution. Deep Think combines scientific knowledge with engineering utility to support practical applications.
Applications by Early Testers
- Mathematics Research: Lisa Carbone, a mathematician at Rutgers University, used Deep Think to review advanced physics mathematics papers. The system identified a subtle logical flaw that had previously gone undetected in human peer review.
- Materials Science: The Wang Lab at Duke University applied Deep Think to optimize complex crystal growth for semiconductor research. Deep Think successfully designed a method for growing thin films larger than 100 μm, achieving targets that prior methods struggled to reach.
- Engineering Design: Anupam Pathak, R&D lead at Google’s Platforms and Devices division and former CEO of Liftware, tested Deep Think to accelerate the design of physical components.
Performance in Mathematics and Algorithms
Specialized versions of Deep Think have previously achieved top-level results in reasoning competitions. The updated version continues this trend, reaching benchmarks in mathematics and programming:
- Humanity’s Last Exam: 48.4% (without tools)
- ARC-AGI-2: 84.6%, verified by the ARC Prize Foundation
- Codeforces: Elo 3455 in competitive programming
- International Math Olympiad 2025: Gold-medal level performance
Scientific and Engineering Applications
Gemini 3 Deep Think demonstrates capability across scientific domains such as physics and chemistry, achieving gold-level results in the written sections of the 2025 International Physics Olympiad and Chemistry Olympiad, and scoring 50.5% on the CMT-Benchmark in advanced theoretical physics.
In applied settings, Deep Think supports research and engineering workflows by helping researchers interpret complex datasets and enabling engineers to model physical systems through code, targeting use cases that require advanced reasoning and structured data analysis.
Availability
- Gemini App: Accessible to Google AI Ultra subscribers
- Gemini API: Available through an Early Access Program for researchers, engineers, and enterprises.