Google DeepMind is currently testing an internal AI model named Mathematica, designed for advanced mathematical problem-solving. This experimental system is a new branch of the Deep Think architecture and is based on the Gemini 3.1 Pro iteration.
The model utilizes parallel-thinking strategies derived from previous projects, including the Aletheia research system. During its reasoning process, Mathematica generates inner monologues that include exclamations of excitement in multiple languages, such as English and Chinese.
The underlying Gemini Deep Think technology previously achieved a gold-medal standard at the 2025 International Mathematical Olympiad. It successfully solved five out of six problems, securing 35 out of 42 possible points.
The development of this model occurs alongside organizational changes at DeepMind, following a restructuring in August 2026 where Demis Hassabis transitioned to the role of chairman. Currently, there is no public release date for the project.
