Full marks at the International Mathematical Olympiad have long been reserved for an elite handful of young human competitors each year, but that exclusivity has now been broken by machines. Two Chinese technology companies, Huawei and Xiaohongshu, announced that their respective AI systems each achieved perfect scores on this month’s IMO questions, marking the first time artificial intelligence has matched that feat under the competition’s official judging process.
The two firms were the first to come forward with their results, though industry watchers expect other AI labs around the world to reveal similar scores in the days ahead. Xiaohongshu, known internationally as RedNote, described the achievement as a significant breakthrough, noting in a statement that no large language model had previously managed a perfect score through the IMO’s formal evaluation.
The result represents a rapid leap forward. Just last year, AI systems developed by companies including Google and OpenAI reached gold-medal level performance for the first time, yet still fell short of the five human contestants who achieved perfect scores in that competition.
This year’s contest, held in Shanghai, drew 666 participants from countries around the world, all of them under the age of 20 as required by IMO rules. Among that large field, only seven human contestants managed to answer every question correctly, according to the official scoreboard, a scarcity that underscores just how difficult a clean sweep typically is.
Huawei detailed its own entry, an AI system called Celia, saying the model demonstrated well-rounded problem-solving ability across multiple areas of mathematics. Xiaohongshu, meanwhile, revealed that its model, dots-note-3.0, was competing at the IMO for the first time this year.
Both companies emphasized the integrity of the testing process. Tech firms received the competition problems only after human contestants had already completed the exam, and were given a fixed window of time to submit their AI-generated solutions. Xiaohongshu stressed that human intervention was strictly forbidden throughout testing, with all AI submissions sent directly to IMO organizers for independent grading.
The pace of progress in this area has been striking. At the 2024 Olympiad, Google’s AI system reached only silver-medal standard, taking two to three days to work through four of the six problems, a stark contrast to this year’s near-instant perfect scores.
Independent testing outside the official competition suggests the trend extends well beyond these two companies. Deedy Das, a partner at Menlo Ventures, said he ran this year’s IMO questions through four leading AI models, from OpenAI, Anthropic, the startup Axiom Math, and Moonshot AI’s Kimi K3, and found that all four scored a perfect 42 out of 42 points. Reflecting on the results, Das suggested that AI capability in mathematics has now moved decisively beyond what Olympiad-level problems can meaningfully test.
