Key Highlights

  • Huawei and Xiaohongshu each reported their AI systems achieved 100 percent on this year's International Mathematical Olympiad questions.
  • Just seven of 666 human contestants achieved full marks at the 2026 contest held in Shanghai.
  • No large language model had previously achieved a perfect score under the IMO's official judging process.
  • Four other AI models, including from OpenAI and Anthropic, separately scored 42 out of 42 when independently tested on the same questions.

A First for AI at the World's Top Mathematics Contest

A perfect result at the International Mathematical Olympiad has historically been reserved for a handful of exceptionally talented young mathematicians each year, but for the first time, artificial intelligence systems have also achieved full marks. Chinese technology companies Huawei and Xiaohongshu said their AI models each scored 100 percent on the questions given to human contestants at this year's IMO, held in Shanghai from July 10 to 21, becoming the first AI labs to publicly announce such results, with others potentially following suit in subsequent days.

Xiaohongshu described the achievement as extremely challenging to attain, noting that no large language model, the underlying technology powering AI chatbot tools, had previously achieved a perfect score under the IMO's official judging process. Just seven of the 666 human contestants competing this year achieved full marks, according to the official IMO scoreboard, with all participants required to be under the age of 20.

A Sharp Acceleration From Recent Years' Results

The achievement marks a notable acceleration from the prior year, when models developed by companies including Google and OpenAI reached gold-level scores for the first time but still fell short of matching the five human contestants who achieved perfect scores in 2025. Huawei said its AI system, called Celia, demonstrated comprehensive problem-solving capabilities across several mathematical fields, while Xiaohongshu, known internationally as RedNote, said its model, dots-note-3.0, was competing at the IMO for the first time this year.

Technology firms participating in this informal AI benchmarking exercise received the IMO problems only after human contestants had already taken the exam, submitting their AI-generated solutions within a specified time limit. Xiaohongshu said any form of human intervention was strictly prohibited during testing, with AI solutions submitted directly to IMO organisers for grading under the same process applied to human contestants.

Progress Has Accelerated Rapidly Across the Industry

The pace of improvement in AI mathematical reasoning has been notably fast. At the 2024 IMO, Google's model achieved only a silver-medal score, solving four of six problems over a period of two to three days, a considerably slower and less complete performance than this year's reported results. A partner at US investment firm Menlo Ventures said he separately administered this year's IMO questions to four leading AI models from US labs OpenAI and Anthropic, startup Axiom Math, and Moonshot AI's Kimi K3 from China, reporting that all four scored a perfect 42 out of 42 points, suggesting the achievement may not be confined to the two companies that first announced results.

Conclusion

The emergence of multiple AI systems achieving perfect scores at one of mathematics' most demanding competitions signals a meaningful shift in the pace of AI reasoning capability development, with models from both Chinese and American labs apparently reaching this threshold in close succession. Whether this milestone reflects a broader step change in AI mathematical reasoning capability, or a more narrow achievement specific to well-defined competition-style problems, may become clearer as additional labs report their own results and as these systems face novel mathematical challenges beyond established contest formats.