HUMANS OUTPERFORM AI IN INTERNATIONAL MATH OLYMPIAD, MACHINES CATCHING UP
In a remarkable display of human intellectual prowess, students participating in the 2025 International Mathematical Olympiad (IMO) have once again outperformed artificial intelligence systems developed to solve complex mathematical problems. Held in Tokyo, Japan, this year’s competition saw 112 countries send their brightest young minds to tackle challenging math problems that test ingenuity, creativity, and deep understanding — areas where humans still maintain an edge over machines.
Despite rapid advances in AI and machine learning, particularly in natural language processing and pattern recognition, solving abstract mathematical problems remains a domain where human intuition continues to lead. AI systems submitted by research institutions and companies such as DeepMind, OpenAI, and Google Brain were invited to participate in parallel, non-competitive trials designed to evaluate the machines’ capabilities alongside human participants.
The results were clear: while AI systems successfully solved many problems from previous IMO years and even made significant progress on selected 2025 problems, they still lagged behind top-performing human contestants, particularly in areas requiring long-term reasoning, abstract thinking, and unconventional problem-solving methods. The gold medal winner, 17-year-old Yifan Zhou from China, achieved a near-perfect score with elegant, highly original solutions that impressed judges worldwide.
“Machines are faster and more accurate in computation, but human competitors excel in adaptability, insight, and creative leaps,” said Dr. Aiko Tanaka, lead organizer of the 2025 IMO. “These problems aren’t just about calculation — they demand intuition and lateral thinking, which AI still struggles to replicate.”
AI programs involved in the trials were trained on massive datasets of solved math problems and used advanced symbolic reasoning models. Some even leveraged large language models fine-tuned for mathematical logic. However, when confronted with unfamiliar structures or the need to invent new approaches — something frequently required at the IMO — AI performance dropped off significantly.
Nevertheless, experts warn that the gap is closing. DeepMind’s AlphaMath system, for instance, solved four of the six problems and produced partial solutions for a fifth — a significant leap from its performance just a year ago. “It’s not a matter of if, but when,” said Professor Marcus Liu of MIT, an advisor to one of the AI teams. “Within a decade, AI may consistently match or exceed the top 10% of human contestants.”
Competitors at the Olympiad welcomed the challenge. “It’s inspiring, not intimidating,” said Zoe Kim, a silver medalist from South Korea. “If AI pushes the boundaries, it also pushes us to think deeper and go further.”
The continued inclusion of AI in these academic contests has sparked debate in the mathematical community. Some see it as a valuable research tool and a benchmark for AI development, while others caution against overestimating its capabilities in fields rooted in creativity and abstract reasoning.
For now, humans remain at the top of the mathematical mountain — but the steady climb of artificial intelligence suggests that a new summit may soon be within reach. The 2025 IMO served not only as a celebration of human brilliance but also a glimpse into a future where humans and machines may one day compete — or collaborate — in the pursuit of mathematical discovery.