Experimental model by OpenAI wins gold at Math Olympiad

OpenAI has achieved a remarkable milestone by attaining “gold medal-level performance” at the International Math Olympiad, highlighting the rapid advancements in the field of artificial intelligence. According to Alexander Wei, a research scientist at OpenAI focusing on large language models (LLMs) and reasoning, an experimental research model successfully tackled this “longstanding grand challenge in AI,” showcasing the potential of AI in complex problem-solving.

Wei revealed that an unreleased model from OpenAI managed to solve five out of six challenging problems presented at this prestigious math competition, accumulating an impressive 35 out of 42 points overall. The International Math Olympiad (IMO) is a premier event where countries send up to six talented students to confront extremely difficult algebra and pre-calculus problems. Although these problems may appear straightforward at first glance, they typically require a significant degree of creativity and critical thinking to achieve the highest possible scores. In this year’s event, only 67 of the 630 contestants were awarded gold medals, representing a mere 10 percent of participants.

While AI systems are adept at handling complex datasets and performing repetitive tasks, they often struggle with challenges that necessitate higher levels of creativity or intricate decision-making processes. However, OpenAI’s latest model proved capable of addressing complicated mathematical problems with a reasoning ability that closely resembles that of human mathematicians, marking a significant breakthrough in the application of AI.

“By achieving this, we have developed a model that can construct complex, watertight arguments comparable to those produced by human mathematicians,” Wei stated in his post on X. Both Wei and Sam Altman, the CEO of OpenAI, emphasized that the company does not anticipate releasing a model with this exceptional mathematical prowess for several months. Consequently, the forthcoming GPT-5 is expected to be an enhanced version of its predecessor, although it is unlikely to possess the same extraordinary capabilities demonstrated in the IMO competition.