Chinese AI models claim full marks at top maths contest

Two Chinese technology companies, Huawei and Xiaohongshu, said their artificial intelligence models scored 100 percent on the same questions given to human contestants at this month's International Mathematical Olympiad in Shanghai.

The companies were the first AI labs to announce scores from this year's contest, and other labs may follow in the coming days. Xiaohongshu said in a statement that achieving a perfect score at the IMO is extremely difficult and said no large language model had previously achieved that result under the IMO's official judging process.

In 2025, models from companies including Google and OpenAI reached gold-medal scores for the first time, but they did not match the five human contestants who earned perfect scores. This year, 666 contestants from different countries took part in the IMO, and seven achieved full marks, according to the official scoreboard. Contestants must be under 20.

Huawei said on Wednesday that its AI system, Celia, had shown broad problem-solving ability across several mathematical fields. Xiaohongshu, also known in English as RedNote, said its dots-note-3.0 model entered the contest for the first time this year.

Technology firms received the IMO problems only after human contestants had taken the exam, and their answers had to be submitted within a set time limit. Xiaohongshu said human intervention was strictly barred during testing and that the AI solutions were sent to IMO organisers for grading.

AI results at the contest have advanced quickly. At the 2024 IMO, Google achieved a silver-medal score after solving four of six problems over two to three days.

Deedy Das, a partner at the US investment firm Menlo Ventures, said on LinkedIn that he gave this year's IMO questions to four advanced AI models. He said all four, from OpenAI, Anthropic, Axiom Math and Moonshot AI's Kimi K3, scored 42 points out of 42.