Xiaohongshu Large Model Achieves Full Marks in IMO: China's AI Makes Its Debut as a Math Olympiad Champion, the Solution to Question 3 Delights the Winner with Its Elegance

The results of the 67th International Mathematical Olympiad (IMO2026) have recently been announced, with the Xiaohongshu large model dots-note-3.0 achieving a perfect score by answering all six questions correctly, winning the gold medal. This marks the first time a Chinese large model has received official gold-level certification from the IMO, and it is also the second global large model to achieve this feat after Google's Gemini — and it did so with a perfect score, while Google previously solved only 5 out of 6 questions.

IMO stands for International Mathematical Olympiad, an annual competition that gathers top high school students worldwide. Half of today's Fields Medal winners, including Tao Te-Xun, come from this competition. The contest takes place over two days, with participants solving three problems each day within 4.5 hours. The problems cover algebra, combinatorics, geometry, and number theory, with each question worth 7 points, totaling 42 points. Participants must provide logical, complete proofs that can be reviewed step by step, which presents a significant challenge for large models.

Google's Gemini provides a reference point: in its first attempt in 2024, it earned a silver medal with 28 points, and required translating the questions into formal languages like Lean, with some problems taking several days to solve; in 2025, Gemini Deep Think completed the proofs using natural language end-to-end, solving five questions within 4.5 hours and earning a gold medal. Math competitions are considered a test of a large model's intelligence and reasoning capabilities, and IMO has a unique advantage compared to regular benchmarks — unpredictability. The competition problems are created by experts and kept strictly confidential, so models cannot train specifically in advance but must rely on real-time understanding, exploration, and rigorous reasoning.

The gold medal line for this year's IMO was 29 points, a decrease of 6 points from last year's 35 points, indicating that the overall difficulty of the competition was significantly higher than last year. A score of 29 means solving four problems completely and scoring one point from the remaining two to enter the gold medal range, while the Xiaohongshu large model dots-note-3.0 answered all questions correctly, exceeding the gold medal line by a full 13 points.

The perfect score first proves the stability of the Agentic reasoning system. The model needs to understand natural language conditions, identify key information, propose intermediate conclusions, and organize them into complete proofs. Any misinterpretation or skipped reasoning steps will be directly pointed out by the evaluators. That dots-note-3.0 achieved a perfect score across six different types of problems means its performance was not dependent on luck in any single question. Additionally, the model processes problems directly in natural language, offering a complete loop from problem understanding to answer expression, representing its ability to perform transferable general reasoning.

In the specific problem-solving process, dots-note-3.0 used an agent approach, combining natural language and Python code execution for reasoning and problem-solving, while also leveraging recursive self-criticism to review its own arguments. What truly surprised people was the third question — a combinatorial game problem. The common solution is to transform the original problem into a graph connectivity issue and then establish a proof. However, dots-note-3.0 used induction instead, capturing the most essential structure of the problem and designing an appropriate induction object and proposition.

Li Hanzuo, a two-time CMO gold medalist, commented that it was "correct and elegant, with a tight structure and natural logic, ranking among the more concise and elegant solutions in IMO." Wang Qiantong, another CMO gold medalist, said it was "clear and direct, hitting the essence, with every step logically sound, but it would be difficult for human competitors to think of approaching it from this angle."

For Xiaohongshu, the perfect score in the IMO is an important technical showcase. Previously, the public perception of its technology was mainly focused on content communities, recommendation algorithms, and content understanding, with no publicly authoritative evidence of its position in the field of foundational models. The perfect score in the IMO is different — 42 points are right there, without the need for complex explanations, sufficient to prove that Xiaohongshu now possesses a foundational model capability worthy of serious industry consideration, marking the emergence of a new player to watch in the field of foundational models. It is reported that dots-note-3.0 will be open-sourced soon.