Humans beat AI at the 66th International Mathematical Olympiad

▲ Image made with Meta AI

Monica Mateos

La Jornada Newspaper, Thursday, July 24, 2025, p. 6

Upon hearing the news that artificial intelligence (AI) models from OpenAI and Google DeepMind achieved gold medal-like performance at the 66th International Mathematical Olympiad (IMO) held in Australia, the global mathematical community immediately retorted: yes, but humans beat AI, young pre-university students who still have much to contribute to the future .

None of the models earned the top score of 42, which five students from China, Canada, Japan, and Russia achieved in the prestigious annual competition, in which entrants must be under 20 years old.

Not only that: the computer giants said their AIs scored 35 points. However, at this year's IMO, not counting the five perfect exams, 21 pre-university students from China, the United States, Vietnam, India, Korea, Russia, Austria, Canada, Slovakia, Hong Kong, Israel, Italy, and Mongolia surpassed the cyberscore.

Another statistic in favor of humans: 46 competitors achieved the 35 points claimed by OpenAI and Google (among them Peruvians Renato Gaitán and Sebastián Lozada, the only Latin Americans in that ranking ); and 61 students tackled problem six, which the computer models couldn't solve.

According to AFP, the IMO organizers cautioned that they were unable to verify how much computing power the AI models used or whether human participation was involved.

On behalf of Mexico, two students won silver medals with 28 points each: Takumi Higashida Martínez from Mexico City and Emmanuel Buenrostro Briseño from Jalisco; while the bronze medals went to Héctor Juan Villarreal Corona (27 points) and Mateo Iván Latapí Acosta from Jalisco, as well as Iker Torres Terrazas from Chihuahua (both with 24 points), and Javier Caram Quirós, also from Mexico City, with 22 points.

Beyond the Olympics and medals, mathematicians are betting that truly great things will come when artificial intelligence models help them solve very complex (and, until now, impossible) problems that require millions of steps and procedures.

Some experts even say that we could be close to solving the now legendary Millennium Problems posed by the Clay Mathematics Institute of Cambridge, Massachusetts, United States (CMI), in 2000, which offers a reward of one million dollars each for its solution.

The seven chosen problems, considered by the CMI to be important classical questions that have not been solved in years , are: P vs NP, which asks whether there are problems that can be verified quickly (in polynomial time) and that can also be solved early; the Hodge conjecture, a deep problem about the structure of certain geometric objects called complex algebraic manifolds; the Poincaré conjecture, which shows that the three-dimensional sphere is the only compact simply connected manifold; the Riemann hypothesis, a central problem in number theory involving the distribution of prime numbers; the Navier-Stokes equations, which establishes the existence and smoothness of solutions to these equations describing fluid motion; the Birch-Swinnerton-Dyer conjecture, which relates the algebraic structure of certain elliptic curves to analytic properties; and the mass gap , a problem in quantum physics involving the existence of a mass gap in Yang-Mills theory.

In 2002, the Poincaré conjecture was solved by Russian mathematician Grigory Perelman (St. Petersburg, 1966), and endorsed in 2010 by the IMC. For this achievement, he was also awarded the Fields Medal, considered the Nobel Prize of mathematics. Perelman declined both prizes, saying he didn't want to be on display like an animal in a zoo. "I'm not a hero of mathematics. I'm not even that successful. That's why I don't want everyone watching me ." In 1982, as part of the Soviet team, he won a gold medal at the IMO in Budapest, with a perfect score of 42 points, a medal he did accept.

In October 2024, Mark Zuckerberg's company, Meta, announced that its AI had unraveled Lyapunov functions, a mathematical problem that has puzzled experts for more than 130 years and offers key tools for analyzing the stability of dynamical systems.

In this sense, AI will be welcomed to enhance research in an area where reasoning and creativity will remain largely human, the mathematical community agrees.