The latest experimental iteration of a reasoning AI focuses on high level mathematical capabilities using an explicit chain of thought method. This specialized model, called Gemini DeepThink Mathematica, generates internal logic sequences to solve complex problems and is currently in a testing phase.

The math focused version of the DeepThink system aims to improve performance on the Millennium bench, a benchmark for AI reasoning. Examples of the model's internal processing show the "raw thoughts" the system creates before providing a final answer, indicating a targeted effort by Google to compete in advanced mathematics.