Google upgrades Gemini 3 Deep Think AI model
Google has completed a major upgrade of its reasoning model, Gemini 3 Deep Think. The updated version reportedly delivers significant improvements in mathematics, physics, chemistry, and programming, enabling engineers to model physical objects for 3D printing.
Gemini 3 Deep Think is positioned as a specialized reasoning mode designed to push the boundaries of intelligence and tackle modern challenges in science, research, and engineering.
The project team stated that the updated version was developed in close collaboration with scientists and researchers and is aimed at solving complex research problems “where challenges often lack clear constraints or a single correct answer, and where data may be incomplete or unsystematic.”
Competition among AI models intensifies
Among Gemini 3 Deep Think’s achievements is setting a new benchmark of 48.4% (without tools) on the Humanity’s Last Exam test, which is designed to assess the limits of modern AI models. The model also outperformed several competitors on ARC-AGI-2, scoring 84.6%, as confirmed by the ARC Prize Foundation.

Gemini 3 Deep Think achieved a rating of 3455 on Codeforces, a competitive programming platform, and demonstrates advanced proficiency in scientific disciplines such as physics and chemistry.
The new Deep Think version is now available in the Gemini app for Google AI Ultra subscribers. It will also be offered for the first time through the Gemini API to selected researchers, engineers, and enterprises as part of an early access program.
The team emphasizes its goal of making Deep Think highly accessible for practical applications and accelerating real-world problem-solving, including 3D printing and other advanced engineering fields.
As we wrote, Google Cloud expands European reach with Liberty Global AI partnership
- Forex
- Crypto