Xiaohongshu's AI Model Achieves First Perfect Score at International Mathematical Olympiad
Miles Bennett
Xiaohongshu's AI model dots-note-3.0 scored a perfect 42 out of 42 at this year's IMO — the first AI model ever to achieve a perfect score, marking a leap from gold-medal level to flawless in just one year.
What does a perfect score actually prove?
dots-note-3.0 solved all six problems at this year's IMO, earning a perfect 42 points — the first AI model in history to do so.
This means → the model didn't just produce correct answers; it wrote complete mathematical proofs rigorous enough to satisfy IMO judges, who grade on logical structure, not final numbers alone.
The IMO wrapped up Monday in Shanghai. Held annually since 1959, it is widely regarded as the world's most authoritative mathematics competition.
How does this compare to last year's gold medalists?
At last year's IMO, Google DeepMind and OpenAI each scored 35 out of 42 — a gold-medal performance.
dots-note-3.0 pushed that to a perfect 42. In plain terms = in just one year, AI math reasoning jumped from "top competitor" to "flawless."
This reflects a pace of improvement that far outstrips expectations — last year's gold-medal benchmark is already obsolete.
What is dots-note-3.0?
It is the smallest-parameter model in Xiaohongshu's dots3 family, still in its testing phase.
This means → the perfect score came from the lightest version in the lineup. Two larger siblings — jazz and aria — are optimized for different use cases and compute budgets, and their potential remains untested at IMO level.
Xiaohongshu said it will open-source dots-note-3.0 "in the near future" but disclosed no specific timeline.
What does this mean for the AI industry?
The jump from 35 to 42 happened in a single year. This means → existing AI reasoning benchmarks are being saturated at a rate that renders them less useful as differentiators.
In plain terms = the exam used to measure how "smart" an AI is has been aced — the industry now needs harder tests.
It also means a company best known for social content has posted a single-event result that surpasses both Google and OpenAI in mathematical reasoning.
Content is for reference only, not financial advice.