OpenAI and Google outdo the mathletes, but not each other

In a remarkable display of artificial intelligence prowess, both OpenAI and Google DeepMind secured gold medal scores at the 2025 International Math Olympiad (IMO), a prestigious competition known for challenging high school students. This achievement, announced by both companies recently, highlights the rapid advancements in AI technologies while illustrating the close competition between these two industry giants. The IMO is not just a test of mathematical skill but also serves as a benchmark for AI capabilities, particularly since many AI researchers have backgrounds in competitive mathematics. OpenAI and Google participated with informal AI systems that could directly process questions and generate proof-based answers without needing human translation, a significant upgrade from Google's previous year's silver medal performance, which relied on a formal system requiring human assistance. Both companies reported that their AI models successfully answered five out of six questions posed during the competition, outperforming a majority of high school participants and surpassing Google's own performance from the prior year. Researchers from both organizations expressed that these accomplishments mark significant breakthroughs in AI reasoning, especially in non-verifiable domains where ambiguity is prevalent. However, the competition has sparked a debate regarding the timing and accuracy of OpenAI's announcement. Google DeepMind's leadership took to social media to criticize OpenAI for prematurely revealing its results, suggesting that OpenAI did not wait for an official evaluation by IMO before making its claims. Thang Luong, a senior researcher at Google DeepMind, emphasized the importance of respecting the students' achievements and following proper evaluation protocols, as they waited for official confirmation from IMO before announcing their results. In contrast, Noam Brown, an OpenAI researcher involved in the IMO project, explained that the organization had been approached by IMO for a formal competition but opted to focus on natural language systems instead. OpenAI claims it engaged third-party evaluators to assess its model's performance accurately and adhered to the IMO's request for a delayed announcement until after the award ceremony. While Google’s approach followed a more structured evaluation process, the broader narrative underscores the rapid improvement of AI models across leading labs. The IMO saw some of the brightest students from around the globe, and the performance of OpenAI and Google’s AIs reflects a significant leap in AI capabilities, with both organizations vying to maintain a competitive edge as OpenAI prepares for the anticipated release of GPT-5 in the coming months.