DeepMind's AlphaGeometry2 AI Outperforms Olympiad Gold Medalists in Geometry

The upgraded AI system solved 84% of International Mathematical Olympiad geometry problems, surpassing the average human gold-medalist score.

AlphaGeometry2, developed by Google DeepMind, achieved a solving rate of 84% on past International Mathematical Olympiad (IMO) geometry problems, surpassing the average gold-medalist score of 40.9 out of 50 problems.
This marks a significant improvement from its predecessor, AlphaGeometry, which solved only 54% of IMO geometry problems a year ago.
The system combines Google's Gemini large language model with a neuro-symbolic engine, enabling it to perform rigorous deductive reasoning and solve complex geometric proofs.
AlphaGeometry2 was trained on a dataset of over 300 million synthetic theorems and proofs, designed to enhance its logical reasoning and problem-solving capabilities.
Researchers view this achievement as a step toward developing more advanced AI systems capable of tackling broader mathematical and logical challenges.