The results of the 67th International Mathematical Olympiad (IMO 2026) have been announced: Xiaohongshu’s large model, dots-note-3.0, achieved a perfect score by correctly solving all six problems, earning a gold medal. This marks the first time a Chinese large model has received official IMO gold medal certification, and the second large model globally to achieve this feat after Google’s Gemini—while Gemini previously solved only 5 out of 6 problems. The IMO is regarded as a benchmark for evaluating the intelligence and reasoning capabilities of large models, requiring participants to provide logically complete proofs—a task of extreme difficulty. This year’s gold medal cutoff was 29 points; dots-note-3.0 scored 13 points above the cutoff. Dual CMO gold medalists praised its solutions as “correct and elegant, with compact structure and natural logic.” This perfect score demonstrates that Xiaohongshu has developed transferable general reasoning capabilities, introducing a noteworthy new player in the field of foundational models.Article author and source: Xiaohongshu
The results of the 67th International Mathematical Olympiad (IMO 2026) have been announced, with Xiaohongshu’s large model, dots-note-3.0, earning a gold medal by correctly answering all six questions with a perfect score. This marks the first time a Chinese large model has received official IMO gold-level certification, and it is the second large model globally to achieve this result after Google’s Gemini—achieving a perfect score, whereas Gemini previously answered only 5 out of 6 questions correctly.
IMO is the abbreviation for the International Mathematical Olympiad, an annual competition that brings together the world's top high school students; half of today's Fields Medalists, including Terence Tao, are alumni of this event. The competition spans two days, with participants solving three problems each day within 4.5 hours, covering algebra, combinatorics, geometry, and number theory. Each problem is worth 7 points, for a maximum score of 42. Contestants must provide logically complete proofs that can be systematically verified—a challenge of extreme difficulty for large language models.
Google Gemini’s journey at the IMO serves as a reference: In its first attempt in 2024, it earned a silver medal with only 28 points, requiring problems to be first translated into formal languages like Lean, with some problems taking days to solve; in 2025, Gemini Deep Think solved all five problems in 4.5 hours using natural language end-to-end, earning a gold medal. Mathematical competitions are regarded as a litmus test for large models’ intelligence and reasoning capabilities, and the IMO has a unique advantage over conventional benchmarks—its unpredictability. Each year’s problems are crafted by experts and kept strictly confidential, preventing models from training on them in advance; instead, they must rely solely on real-time understanding, exploration, and rigorous reasoning.
The gold medal cutoff for this year's IMO is 29 points, a decrease of 6 points from last year's 35 points, indicating that the overall difficulty of this year's exam is significantly higher. A score of 29 points means solving four problems completely and earning one additional point from the remaining two problems is sufficient to qualify for a gold medal. However, the Xiao Hong Shu large model dots-note-3.0 answered all questions correctly, exceeding the gold medal cutoff by a full 13 points.
A perfect score first demonstrates the stability of the Agentic reasoning system. The model must comprehend natural language conditions, identify key information, generate intermediate conclusions, and organize them into a complete proof—any misinterpretation of conditions or skipped logical steps will be immediately flagged by reviewers. dots-note-3.0 achieving a perfect score across six distinct problem types indicates that its performance is not due to accidental insight on any single question. Furthermore, the model processes inputs end-to-end in natural language, enabling a complete closed loop from problem understanding to answer expression, demonstrating its transferable, general-purpose reasoning capability.
During the problem-solving process, dots-note-3.0 employs an agent-based approach, combining natural language with Python code execution to reason through solutions, while leveraging recursive self-criticism to evaluate its own arguments. What is truly impressive is the third question—a combinatorial game theory problem. The conventional approach involves transforming the problem into a graph connectivity issue and constructing a proof accordingly; however, dots-note-3.0 instead applies mathematical induction, identifying the most fundamental structure of the problem and designing an appropriately precise inductive hypothesis and object.
Liu Hanzuo, a gold medalist at the CMO, praised it as "correct and elegant, with a compact structure and natural logic, making it one of the more concise and graceful solutions even among IMO-level answers." Wang Qiantong, another CMO gold medalist, noted that it is "clear and concise, directly addressing the core issue—each step flows logically, yet it’s difficult for human competitors to think of approaching the problem from this angle."
For Xiaohongshu, IMO's perfect score is a significant technological showcase. Previously, external perceptions of its technology were centered on content communities, recommendation algorithms, and content understanding, with no publicly recognized validation of its capabilities in foundational models. IMO’s perfect score is different—42 points speak for themselves, without needing complex explanations, proving that Xiaohongshu now possesses foundational model capabilities worthy of serious industry attention, introducing a new player to watch in the foundational model space. It is reported that dots-note-3.0 will soon be open-sourced.