Two Chinese tech giants, Huawei and Xiaohongshu, announced that their AI models both scored 100 percent on the problems presented to human participants at this month’s International Mathematical Olympiad (IMO).
They were the first AI laboratories to disclose their achievements, with others worldwide likely to follow in the upcoming days.
“We are thrilled with this achievement, as obtaining a perfect score at the IMO is incredibly difficult,” Xiaohongshu stated this week.
“Until now, no large language model had ever secured a perfect score through the IMO’s official judging process,” the company continued, highlighting the technology that powers AI applications like chatbots.
In 2025, models from companies like Google and OpenAI attained gold-level scores for the first time—but they could not replicate the success of five human participants who achieved full marks.
‘Frontier of AI’
This year, 666 contestants from various nations participated in the event held in Shanghai.
According to the official IMO leaderboard, only seven participants achieved perfect scores.
Competitors at the IMO must be under 20 years old and the event rotates its location each year.
Huawei announced on Wednesday that its AI system “Celia” demonstrated “comprehensive problem-solving abilities” across various mathematical domains.
Xiaohongshu, also known as RedNote in English, revealed that its model “dots-note-3.0” entered the IMO for the first time this year.
Tech companies received the IMO problems only after human participants had completed the examination, with their solutions submitted under a defined time constraint.
“During the testing phase, no form of human assistance was allowed,” Xiaohongshu explained, stating that the AI-generated solutions were provided to IMO organizers for evaluation.
The advancement of AI has been rapid: at the 2024 IMO, Google earned a silver medal, successfully solving four out of the six problems over a span of two to three days.
Deedy Das, a partner at the US-based investment firm Menlo Ventures, shared on LinkedIn that he had presented this year’s IMO questions to four state-of-the-art AI models.
All four—developed by leading US firms OpenAI and Anthropic, the startup Axiom Math, and “Kimi K3” from China’s Moonshot AI—achieved full marks of 42 points out of 42, he noted.
“The frontier of AI has officially surpassed IMO mathematics,” Das remarked.
