For the first time an artificial intelligence (AI) system has earned a perfect score at the International Mathematical Olympiad, South China Morning Post reports. The model, called dots-note-3.0, was built by the Chinese company RedNote, known in China as Xiaohongshu. It solved all six problems and received the maximum 42 points at this year’s contest, which finished in Shanghai. RedNote has said it plans to make the model open-source, meaning the underlying code will be freely available for others to study and improve, though no exact date has been given. Dots-note-3.0 is the lightest version in the company’s dots3 series; two larger models named jazz and aria are designed for heavier computing needs.
The International Mathematical Olympiad has been held every year since 1959 and is considered the leading contest of its kind. Competitors must supply full written proofs that contain no logical gaps or errors. Simply stating a final numerical answer is not enough. In previous years artificial intelligence systems reached high but incomplete marks. Last year systems from Google DeepMind and OpenAI each scored 35 out of 42. In 2024 another Google system scored 28, yet it needed human specialists to rewrite the problems into a specialised formal language before the machine could begin work.
How the system worked
This year RedNote’s model read the original contest documents directly and completed every step without any human intervention. Official rules forbid hints, corrections or extra help of any kind. The system used an agentic framework, a design that lets the program plan its own actions, switch between ordinary language reasoning and the execution of Python programming code, then test and revise its work. According to the company the model repeatedly examined its own reasoning, questioned its assumptions and corrected mistakes through a multi-step self-review loop before submitting final answers. The Chinese contestants also performed strongly, with several individuals achieving full marks.