A Perfect Score on the World Stage
The world of mathematics is buzzing with news from the most recent International Mathematical Olympiad (IMO), the planet's most prestigious and difficult maths competition for high school students. For the first time, an AI model has achieved a perfect
score. Chinese tech firms Huawei and Xiaohongshu announced their models solved all six incredibly complex problems, matching the top human contestants. This marks a monumental leap from just a year or two ago, when AIs from labs like Google DeepMind were achieving impressive, but imperfect, gold and silver-medal-level scores. Getting a perfect score at the IMO is exceptionally rare even for humans. It requires not just calculation, but a deep, intuitive creativity to construct rigorous, logical proofs—something long considered a uniquely human trait.
More Than Just Brute Force
So how did AI make this jump? The secret lies in moving beyond brute-force computation. Early AI systems struggled with the kind of abstract reasoning needed for Olympiad-level geometry or number theory. Today’s pioneering models, like Google DeepMind's AlphaGeometry, use a powerful hybrid approach. They combine a neural network, which functions like human intuition to find promising paths and suggest creative constructions, with a symbolic engine that rigorously checks the logic of each step. This neuro-symbolic method allows the AI to explore possibilities a human might miss while ensuring its final answer is a logically sound, verifiable proof. To overcome the lack of training data at this elite level, researchers also developed ways to generate hundreds of millions of synthetic practice problems, letting the AI train itself without needing human examples.
From Student Competitions to Unsolved Mysteries
While acing the IMO is a headline-grabbing achievement, the true goal is far more ambitious. These AI systems are being designed to tackle genuinely unsolved problems in mathematics—conjectures that have stumped researchers for decades. In May 2026, an AI from OpenAI made significant progress on the 80-year-old Erdős unit distance problem, discovering a new way of looking at the puzzle that went against long-held human assumptions. Similarly, Google's AlphaEvolve system has been used to chip away at notoriously difficult problems in a field called Ramsey theory. These breakthroughs demonstrate that AI isn't just learning what we already know; it's capable of generating novel insights that can push the boundaries of mathematical knowledge.
A New Partner for Mathematicians
The rise of mathematical AI doesn't spell the end of human mathematicians. Instead, it signals the beginning of a powerful collaboration. Researchers envision AI as a tireless, creative assistant that can explore countless avenues, test hypotheses, and manage complex calculations, freeing up human experts to focus on the bigger picture and guide the overall direction of research. Leading institutions are already embracing this future. Google has launched an "AI for Math Initiative," partnering with universities to explore how these tools can accelerate discovery. The consensus is that AI will become an indispensable tool, augmenting human intuition rather than replacing it. It can explore paths that a person might dismiss as too time-consuming, uncovering surprising connections and solutions.
