The China AI model scored full marks in the International Mathematical Olympiad.

In July 2026, the 67th International Mathematical Olympiad (IMO) concluded in Shanghai. The event attracted approximately 685 participating students from over 120 countries and regions, with questions covering four fields: algebra, geometry, number theory, and combinatorics, each worth 7 points out of a total of 42 points. However, what truly stunned the global mathematics and AI communities was an AI model developed by China’s social media platform Rednote.

On July 22, Xiaohongshu officially announced that its developed dots-note-3.0 model answered all six questions correctly in the 2026 International Mathematical Olympiad (IMO), achieving a perfect score of 42 points—the first time an AI model has scored full marks in the competition.

The significance of this achievement must be assessed within its historical context. Established in 1959, the International Mathematical Olympiad (IMO) is widely recognized as the world’s premier mathematics competition. Unlike traditional math tests that merely require final answers, the IMO demands rigorous proof processes with no logical errors or missing conditions—precisely the greatest weakness of previous AI systems. While AI systems demonstrated notable performance in the IMO, they never reached the top. In 2024, Google DeepMind’s AlphaProof and AlphaGeometry 2 scored 28 points, equivalent to silver medal level, but required human experts to manually translate problems into logical language for AI processing. By 2025, both Google DeepMind and OpenAI achieved gold medal-level scores of 35 points each.

DOTS-note 3.0 remains in the testing phase and represents the most lightweight model in the DOTS3 series to date. The series also includes two larger-scale versions—Jazz and Aria. Notably, this year’s Xiaohongshu model directly utilized raw mathematical competition files without requiring manual translation of problems into specialized programming languages. DOTS-note 3.0 employs an “agent” framework that integrates natural language reasoning with Python programming execution to draft, test, and refine its proof processes.

A perfect score at this year’s IMO is extremely rare—among the official results, only seven students out of over 600 achieved a score of 42, accounting for approximately 1%. Dots-note-3.0 secured a perfect score by exceeding the gold medal threshold by 13 points, marking the first time China’s AI has earned an officially certified “perfect score gold medal” at the IMO and making it the world’s first large-scale model to achieve a perfect score in the official IMO grading process. Xiaohongshu announced that it will open-source this model in the “near future.” Notably, Huawei’s Xiao Yi AI Competition Agent also won the gold medal with a perfect score of 42 at this IMO—both Chinese tech companies reaching the top simultaneously further underscores the milestone significance of this achievement.

The full IMO score of dots-note-3.0 is not an isolated achievement of China’s AI, but rather the latest testament to a series of systematic breakthroughs.

At the level of fundamental model capabilities, China’s AI has already demonstrated the strength to compete head-to-head with global leaders. The IMO full score of “42 points” clearly demonstrates that Xiaohongshu possesses foundational model capabilities worthy of serious industry scrutiny. This achievement makes dots-note-3.0 the second AI model to receive an officially certified gold-level rating, following Google DeepMind’s Gemini DeepThink. The key difference lies in the fact that Gemini only solved five out of six problems by 2025, whereas dots-note-3.0 achieved a perfect score.

In terms of technological innovation, China’s AI is forging a unique path distinct from that of Western giants. In the past, AI challenges for IMO heavily relied on human experts—translating problems into logical language was an essential prerequisite step. This year, Xiaohongshu directly processed raw competition files, achieving end-to-end automation from problem input to solution output. The “agent” framework of dots-note-3.0 integrates natural language reasoning with code execution and employs a multi-step self-review mechanism to iteratively correct flawed logic—a architectural approach that competes differently from Google DeepMind’s symbol-based reasoning approach. Huawei’s Xiaoyi AI Competition Agent demonstrates another technical pathway: combining sophisticated problem modeling capabilities, long-range reasoning engines, and tool coordination frameworks. Both approaches ultimately converge, highlighting China’s profound expertise in foundational scientific intelligence.

In terms of open-source ecosystems and industrial vitality, China’s AI is gaining global influence. Xiaohongshu has committed to making dots-note-3.0 open-source, continuing the path taken by Chinese AI companies in promoting global technology dissemination through open-source models. The cumulative global downloads of China’s open-source large-scale models have surpassed 10 billion, ranking first worldwide. According to Hugging Face’s Spring 2026 report, Chinese models accounted for 41% of total model downloads on the platform in the past year, surpassing the United States for the first time. From DeepSeek to Xiaohongshu, from Huawei to Moon Dark Side, multiple Chinese AI companies are achieving cluster-based breakthroughs across various fields.

The perfect score of dots-note-3.0 on the IMO is not a mere accidental technical stroke of genius, but rather the inevitable outcome of China’s systematic institutional investment in the field of AI.

First, China possesses strategic determination rooted in top-level design. From “fully leveraging the advantages of the new national system, adhering to self-reliance and strength, and emphasizing application-oriented approaches” to “combining an effective market with a proactive government,” China’s policies in the field of artificial intelligence exhibit inherent consistency and continuity. By 2026, China will accelerate comprehensive legislation for the healthy development of artificial intelligence, while intensifying efforts to achieve innovative breakthroughs in AI, expedite breakthroughs in key technologies, and advance the construction of ultra-large-scale intelligent computing clusters. This institutional framework, grounded in long-term planning, provides enterprises and research institutions with a predictable R&D environment.

Secondly, China has pursued an open and inclusive path of multilateralism. At the 2026 World Artificial Intelligence Conference, 29 countries signed an agreement, officially establishing the world’s first intergovernmental international organization on artificial intelligence—the World Artificial Intelligence Cooperation Organization (WAICO). China pledged to provide 5,000 specialized AI training opportunities for developing countries over the next five years. Xiaohongshu committed to making its full-scoring AI model open-source, which exemplifies this open-minded approach at the corporate level: technological breakthroughs are not exclusive patents but public goods shared globally.

Third, China adheres to a governance approach that balances development and security. In response to the systemic risks posed by the rapid iteration of AI technologies, China continuously improves relevant laws, regulations, policy frameworks, application standards, and ethical guidelines to ensure that artificial intelligence is safe, reliable, and controllable. This institutional design, which ensures both rapid progress and stable development, provides a solid foundation for the healthy advancement of AI technology.

The United States focuses on policy adjustments and export controls in the field of AI, while China has provided an alternative pathway for AI development through a stable policy framework, sustained foundational investments, and open multilateral cooperation. From achieving full IMO certification to committing to open-source initiatives, from establishing WAICO to supporting capacity building in developing countries—China is crafting its own AI solution through long-term institutional planning.