For the first time, AI models have posted perfect scores at the International Mathematical Olympiad — and the headline result comes not from a frontier lab but from RedNote, the Chinese social-media company behind Xiaohongshu. Its beta model dots-note-3.0 solved all six problems at the 67th IMO in Shanghai for a flawless 42/42, the company announced July 22. Huawei's Celia matched the feat at the same event.
Under official rules
This wasn't self-graded: participating firms received the problems only after the human exam, submitted within a time limit, and had solutions marked by IMO organizers. RedNote says human intervention was strictly prohibited. "Previously, no large language model had ever achieved a perfect score," the company said.
The human bar
Just seven of roughly 666 human contestants earned full marks this year. The AI leap is stark against 2025, when Google DeepMind and OpenAI celebrated gold-medal performances of 35/42 — a threshold that now looks like a stepping stone.
The smallest of three
dots-note-3.0 is the lightest model in RedNote's dots3 family, sitting beneath the larger "jazz" and "aria." It remains in beta — and RedNote says it will be open-sourced "in the near future," without giving a date.
The subtext
A social-commerce app out-scoring the labs that defined mathematical AI a year ago underlines how quickly elite reasoning capability is diffusing across China's ecosystem — and an open-weight release would put Olympiad-grade mathematics in anyone's hands. Claims circulating among investors that OpenAI, Anthropic and Moonshot models also hit perfect scores this year are self-reported rather than IMO-graded — the official scoreboard, for now, belongs to dots-note-3.0 and Celia.