RedNote claims 42 out of 42 at the mathematical olympiad

RedNote's dots team said an internal version of its model dots-note-3.0 solved all six problems of the 2026 International Mathematical Olympiad in Shanghai and received 42 out of 42 in official marking.

The olympiad ran from 10 to 21 July, with the closing ceremony on 20 July in Shanghai. The company says the organising committee invited it and marked the model's work officially, and that only seven contestants reached a perfect score this year. Note is the lightest model in the dots3 family; in an August post the company called the system a branch of dots3-note Preview. On 22 July Huawei announced the same score for its assistant Xiaoyi, which AFP calls Celia: 42 out of 42. Its press page, dated 24 July, says Xiaoyi was specially invited to take part and quotes the olympiad's president, Gregor Dolinar, congratulating Huawei on a perfect score; the page says nothing of RedNote, and those words are given only by Huawei. In 2025 Google DeepMind reported 35 out of 42, which, in its words, the olympiad's coordinators certified. An experimental OpenAI model also got 35 out of 42, but in the company's own evaluation: on 19 July 2025 OpenAI researcher Alexander Wei wrote on X that the model solved five of the six problems, that for each problem three former IMO medallists independently graded its submitted proof and the scores were finalised after unanimous consensus, and that it worked under the contestants' rules, in two 4.5-hour sessions with no tools or internet. The same day the official OpenAI account announced gold medal-level performance; in the posts of that account that were read there is no score, no graders and no conditions, and a page on OpenAI's own site from February 2026 repeats 35/42 in one sentence. The organisers of the 2025 olympiad, in a release of 19 July 2025 on imo2025.au, do not say which AI companies tested models and give no AI score; they write that such companies were invited to a fringe event and privately tested closed-source models on that year's problems. The olympiad's president, Gregor Dolinar, said that the olympiad cannot validate the methods, including the compute used, any human involvement or reproducibility, and that correct proofs, whether by students or by AI models, are valid. Wei's GitHub repository openai-imo-2025-proofs, which GitHub lists as created on 18 July 2025 (UTC), before his posts of 19 July, holds five text files, problem_1 to problem_5, and a README of one sentence saying it hosts the proofs produced by 'our experimental reasoning LLM' during its evaluation on the 2025 olympiad. It has no file for the sixth problem and gives no scores, graders or conditions; the proofs were not graded here, and the repository stands under the researcher's own account, not OpenAI's. What the record does not claim: that the olympiad certified OpenAI's score, or that Wei says who engaged the three graders. The posts read do not say it and name none of them; several posts of his thread were not read in full.