Google rolls out Gemini 3 Deep Think for advanced science and engineering

Google has launched Gemini 3 Deep Think, a major upgrade to its AI system, designed to tackle the toughest problems in science, research, and engineering. The new mode demonstrates exceptional capabilities in solving mathematical and algorithmic problems while providing support for actual applications through physical system modelling and complex data interpretation.

Gemini 3 Deep Think has already achieved top-tier results across some of the most demanding tests for AI. The system achieved a 48.4% score on “Humanity’s Last Exam” and an 84.6% score on ARC-AGI-2 puzzles while achieving gold-medal performance in the 2025 International Math, Physics and Chemistry Olympiads.

Google Gemini 3 Deep Think model 

The system exhibited exceptional programming and algorithm skills through its competitive coding performance, which reached an Elo rating of 3455 on Codeforces.

Deep Think goes beyond academic assessments because it has been developed to solve real-world challenges. Users have the ability to transform basic drawings into 3D-printing-ready models, while they can also analyse scientific data and create virtual models of intricate engineering systems. Google emphasises that the AI is not just about speed or scale but also about enabling researchers and professionals to push the frontiers of innovation.

The updated Deep Think mode is now live in the Gemini app for Google AI Ultra subscribers. For the first time, early access is also available via the Gemini API, allowing selected scientists, engineers, and enterprises to test the AI’s capabilities in their own projects. Interested users can sign up to join the early access programme and explore the AI’s problem-solving potential.