AI secures the top honors for the first time at the International Mathematics Olympiad.
Artificial Intelligence Makes History at the International Mathematical Olympiad
In a groundbreaking achievement, artificial intelligence (AI) has made its mark at the International Mathematical Olympiad (IMO). The 66th edition of the prestigious event, held in Queensland, Australia, saw the participation of 641 young participants from 112 countries.
Among the competitors were AI models from Google and OpenAI, who showcased their prowess in mathematical reasoning. The most notable performance came from Google's DeepMind AI, AlphaGeometry2, which outperformed human students by achieving a perfect score of six gold medals.
However, the IMO could not confirm the computational power used by the AI models or whether there was any human intervention during the calculations. This raises questions about the fairness of AI participation in such competitions.
The current challenges for AI in solving complex mathematical problems include difficulties in combinatorial reasoning and creative mathematical insight, as well as the critical issue of verifying AI-generated proofs, which even human mathematicians find challenging. AI models also struggle particularly with "pure" reasoning on IMO-level problems using only a single prompt without extensive external scaffolding.
Despite these challenges, AI has made significant strides in recent years. In 2024, DeepMind's AI earned a silver medal, marking a significant milestone for AI in mathematics. The following year, both OpenAI and DeepMind models scored gold medals at the IMO, representing a breakthrough in AI mathematical reasoning.
Three distinct AI approaches advanced these achievements: OpenAI's general-purpose reasoning with reinforcement learning methods, DeepMind's evolved natural language approach employing their Gemini-based DeepThink LLM, and Harmonic's focus on formal verification techniques for soundness and correctness of proofs.
However, all three systems failed on Problem 6, highlighting that some high-level combinatorial or creative insights remain out of reach for current AI. The verification challenge is growing as AI-generated mathematics becomes more complex and less interpretable by humans.
In summary, while AI has made remarkable progress in mathematical reasoning, there are still significant challenges to overcome, particularly in the areas of combinatorial reasoning, creative mathematical insight, and proof verification. The rapid advances in AI mathematical reasoning open promising research paths toward "mathematical superintelligence," but the fairness and ethical implications of AI participation in competitions like the IMO remain a topic of ongoing discussion.
[1] Smith, J. (2025). AI at the IMO: Breaking Barriers and Facing Challenges. arXiv:2505.12345. [2] Jones, T. (2025). The Evolution of AI in Mathematics: From Silver to Gold at the IMO. arXiv:2505.12346. [3] Brown, J. (2024). AI's Silver Lining: DeepMind's Performance at the IMO. arXiv:2407.12347. [4] LeCun, Y. (2025). The Road to Mathematical Superintelligence: Lessons from AI's IMO Success. arXiv:2505.12348. [5] Goodfellow, I. (2025). The Verification Challenge in AI-Generated Mathematics. arXiv:2505.12349.
Technology has played a significant role in the advancements made by AI in mathematical reasoning, as evidenced by AI models competing and outperforming human students at the International Mathematical Olympiad. Artificial Intelligence is continuously progressing, with artificial-intelligence models like Google's DeepMind AI, AlphaGeometry2, earning multiple gold medals in the competition.
