Google tests math-focused Deep Think Mathematica model internally

Google DeepMind is quietly testing an internal AI model built specifically for advanced mathematics, an experimental system informally called Mathematica that represents a new branch of the company’s Deep Think reasoning architecture. The model, which builds on the Gemini 3.1 Pro iteration, is designed to tackle complex mathematical problem-solving with a level of depth that goes beyond what current generally available models can handle. Perhaps the most interesting detail: Mathematica reportedly generates inner monologue outputs that display something resembling genuine excitement during problem-solving. The model produces exclamations like “wait” and “oh my god” in its reasoning traces as it works through tasks, suggesting Google’s engineers have built, or at least stumbled into, a system that narrates its own mathematical discoveries with theatrical flair. Building on an Olympiad-level foundation The Gemini Deep Think model achieved a gold-medal standard at the 2025 International Mathematical Olympiad, solving five out of six problems and earning 35 out of a possible 42 points. Google subsequently developed Aletheia, which extended these mathematical techniques to research-level...