Little Red Book's AI Model Achieves Perfect Score at IMO 2026, Surpassing Google's Gemini

Xiaohongshu’s "dots-note-3.0" model has achieved a historic feat in the world of artificial intelligence: a perfect score at the International Mathematical Olympiad (IMO) 2026.

The model scored 42 out of 42 points, securing the gold medal. This makes it the first Chinese model to achieve an official IMO gold level and the second globally, following Google's Gemini. Notably, Google's previous attempt in 2025 yielded a score of 35 (5/6), while this year's gold line was 29.

A New Benchmark in AI Reasoning

The achievement highlights a significant leap in Large Language Model (LLM) capabilities. Unlike previous iterations that required translation into formal mathematical languages, dots-note-3.0 processed natural language questions directly, generating complete, rigorous proofs in English.

"The solution structure was compact and the logic natural," commented Liu Hanzuo, a CMO Gold medalist. "It belongs to a concise and elegant category of proofs."

Wang Qiantong, another CMO Gold medalist, added that the model's approach "cut straight to the essence" of the problems, a level of insight that is difficult for even human competitors to replicate. The model’s ability to handle complex combinatorial problems, particularly in Question 3, was highlighted as a standout example of its innovative reasoning.

Why the IMO Matters

The International Mathematical Olympiad is widely regarded as the premier test of mathematical reasoning for high school students. For AI researchers, it serves as a critical benchmark to measure "Agentic reasoning" and generalization.

The test is unique because it uses unknown problems, ensuring models cannot rely on memorization or pre-training on specific datasets. Participants must understand the problem, explore potential paths, and construct a logical proof under strict time constraints.

Xiaohongshu's Strategic Move

For Xiaohongshu, this victory is more than just a technical milestone; it is a strategic entry into the foundational model arena. Historically perceived as a social media platform focused on content recommendation, Xiaohongshu aims to use open-sourcing its model to demonstrate its prowess in deep reasoning and coding.

By securing the highest possible score, Xiaohongshu provides a tangible, undeniable metric of its technological maturity, signaling to the industry that it is a serious contender in the race for advanced AI capabilities.

Xiaohongshu plans to open-source the model, inviting the global community to further analyze and build upon its reasoning architecture.