AlphaGeometry2: An AI System Solves Olympiad-Level Geometry Problems

Artificial intelligence (AI) is making rapid progress in various fields. A particularly exciting area is mathematics, where AI systems can increasingly solve complex problems. One example is AlphaGeometry2, a system developed by DeepMind, which has now achieved the performance of an average gold medalist in solving Olympiad-level geometry problems.

Improvements over the Previous Model

AlphaGeometry2 represents a significant advancement over its predecessor, AlphaGeometry. The original language has been expanded to handle more challenging tasks. These include problems involving movements of objects, linear equations of angles, ratios, and distances. Through these and other additions, the coverage rate of the AlphaGeometry language in the International Mathematical Olympiads (IMO) from 2000 to 2024 has been increased from 66% to an impressive 88%.

Optimized Search Process and Architecture

In addition to expanding the language, the search process of AlphaGeometry2 has also been significantly improved. The use of the Gemini architecture allows for more powerful language modeling. A novel knowledge-sharing mechanism that combines multiple search trees also contributes to increased efficiency. Additional improvements to the symbolic computation engine and the generation of synthetic data led to a significant increase in the overall solution rate. AlphaGeometry2 now achieves a solution rate of 84% for all geometry problems of the last 25 years, compared to 54% for the previous model.

Successes at the International Mathematical Olympiad (IMO)

AlphaGeometry2 was part of the system that achieved the silver medal standard at the IMO 2024. This underscores the system's potential to solve complex mathematical problems at the highest level.

Automation from Natural Language to Solution

Another promising aspect is the development towards a fully automated system. The goal is to solve geometry problems directly from natural language input. AlphaGeometry2 plays a central role in this and could fundamentally change the way we deal with mathematical problems in the future.

Conclusion

AlphaGeometry2 impressively demonstrates the progress of AI in the field of mathematics. The ability to solve complex geometry problems at the Olympiad level opens up new possibilities for research and application. The further development towards a fully automated system that works directly with natural language input also promises enormous potential for the future.

Bibliography: Chervonyi, Y., Trinh, T. H., Olšák, M., Yang, X., Nguyen, H., Menegali, M., Jung, J., Verma, V., Le, Q. V., & Luong, T. (2025). Gold-medalist Performance in Solving Olympiad Geometry with AlphaGeometry2. arXiv preprint arXiv:2502.03544. DeepMind. (2024). AI solves IMO problems at silver medal level. DeepMind Blog. DeepMind. (2024). AlphaGeometry: An Olympiad-level AI system for geometry. DeepMind Blog. Rao, A. (2024). AlphaGeometry: DeepMind’s AI System Solves Olympiad-Level Geometry Problems. Analytics Vidhya. Trinh, T. H., et al. (2024). Solving olympiad geometry without human demonstrations. Nature, 628(7881), 320-326. Wikipedia contributors. (2025). AlphaGeometry. Wikipedia, The Free Encyclopedia. Wu, K. (2024). A.I. Solves Advanced Math, Challenging Assumptions About Machine Thinking. The New York Times.