In a groundbreaking achievement, Google DeepMind’s AI, AlphaGeometry2, has successfully solved 84% of International Mathematical Olympiad (IMO) geometry problems from 2000 to 2024. This remarkable feat places the AI on par with gold medalists in one of the world’s toughest mathematics competitions—problems that even most university professors struggle to solve.
The research, published on February 5, 2025, details how AlphaGeometry2 vastly improves upon its predecessor by integrating an advanced symbolic reasoning system, the Gemini architecture, and a novel language modeling approach. These enhancements have dramatically boosted the AI’s ability to tackle complex geometric proofs, involving intricate relationships between angles, ratios, and distances.
Previously, the first version of AlphaGeometry could solve only 66% of IMO geometry problems. With its latest upgrades, AlphaGeometry2 has not only increased its success rate to 88% on recent IMO geometry questions but also achieved silver-medal-level performance at IMO 2024—a milestone in AI-driven mathematical reasoning.
Read also: South Korea blocks Chinese AI startup DeepSeek over data privacy concerns
For decades, solving geometry problems at the IMO level has been considered an ultimate test of human intelligence and creativity. AlphaGeometry2’s ability to outperform most human mathematicians in this domain signals a major leap in AI’s problem-solving capabilities. It suggests that AI could soon assist with scientific research, engineering, and theoretical mathematics, potentially revolutionizing the way we approach complex problem-solving.
DeepMind’s next goal is to train AlphaGeometry2 to process and solve geometry problems directly from natural language, making AI-assisted theorem proving even more accessible.
This breakthrough raises an intriguing question: Could AI surpass human intelligence in creative problem-solving? While AlphaGeometry2 is not yet perfect, its rapid progress hints at a future where AI-powered mathematicians could push the boundaries of human knowledge—solving problems we once thought were beyond reach.
