MuduoLLM (Shicheng Wanxiang) is China’s first large language model for basic education that closely follows the knowledge system of the new curriculum standards. It has completed algorithm registration for internet information services with the Cyberspace Administration of China. Since the platform was made available on the AI application supermarket of the Beijing Municipal Education Commission in June last year, it has been adopted by more than 300 primary and secondary schools in Beijing. Liu Xiaohui, director of the information center at Beijing Sanfan Middle School, said that MuduoLLM has improved the efficiency of teachers’ lesson preparation and homework feedback, reducing the routine and repetitive tasks of daily teaching. How exactly does it empower teachers and support classroom instruction? The reporter conducted an exclusive interview with the R&D team to find out.

Starting with Teaching: A Large Language Model Built for Teachers

"Beijing Normal University is the birthplace of modern education and scientific research in China, and we are committed to running a university rooted in the Chinese context. The character 'Shi' (teacher) in MuduoLLM (Shicheng Wanxiang) carries forward the 120‑year traditions of Beijing Normal University, empowers teachers, supports their professional responsibilities, and helps nurture talent across all fields," said Professor Huang Hua, Dean of the School of Artificial Intelligence at Beijing Normal University. As a large language model tailored to basic education, MuduoLLM (Shicheng Wanxiang) is integrated into the entire teaching and learning process, striving to deeply integrate technological empowerment with the true essence of education and to promote the high‑quality development of basic education in the new era.

The model was jointly developed by Beijing Normal University and TAL Education Group, under the specific guidance of the Beijing Municipal Education Commission and with the support of the Beijing Municipal Science and Technology Commission. It is built on a deep understanding of the new curriculum standards and combines the knowledge system of those standards with the Socratic method. Its features include knowledge explanation, heuristic guidance, scenario creation, and competency‑based teaching. It naturally embeds the cultivation of core competencies into the entire process of intelligent tutoring, providing strong support for the daily implementation of the new curriculum standards. More importantly, it effectively addresses practical needs in educational settings, enhancing its utility and professionalism in real‑world teaching.

With millions of teachers across China, how can the model meet their diverse needs? According to Professor Huang, the R&D team has built a structured, multidisciplinary core educational resource library covering textbooks, exam questions, and lesson plans. Drawing on the expertise of teaching and research specialists from Beijing Normal University and TAL Education Group, as well as frontline teachers from various regions, the team has converted practical teaching experience into high‑quality datasets. This ensures that the model produces professional, context‑appropriate outputs that effectively support the daily teaching and learning of both teachers and students.

In particular, when it comes to reducing teachers’ workload, MuduoLLM (Shicheng Wanxiang) helps ease their heavy administrative routines by intelligently generating high‑quality lesson plans and providing heuristic Q&A. It cuts down on tedious tasks and reduces the burden of lesson preparation, allowing teachers to focus more on student development and the fundamental mission of education. At the same time, the model uses heuristic, inquiry‑based dialogue to encourage students to ask questions and explore independently, subtly cultivating scientific thinking and supporting the early development of top‑notch talent.

Beyond Problem‑Solving: Rigorous Checks to Ease Teachers’ Workload

MuduoLLM is more than just a problem‑solving tool; it also focuses on student development. "In response to the real demands of basic education, MuduoLLM (Shicheng Wanxiang) uses technology to improve quality, increase efficiency, and advance educational equity," said Associate Professor Zhu Qiannan, a member of the R&D team.

As an intelligent tool designed to serve educators, how does MuduoLLM prevent the hallucinations common to large language models and ensure rigorous knowledge delivery? First, it defines the scope of knowledge. Relying on a pre‑established educational resource repository, all outputs of the model are strictly confined to materials from authoritative textbooks, rigorously reviewed exam questions, and standard lesson plans. This blocks inappropriate generation or misinformation. Second, MuduoLLM embeds pedagogical theories at the algorithmic level, ensuring that its output aligns with real‑world teaching logic. Third, teaching and research teams from Beijing Normal University and TAL Education Group, together with frontline expert teachers, conduct strict fact‑checking and expert alignment. This combination of defining the knowledge scope and expert verification effectively reduces the risk of model errors and guarantees that all outputs are scientifically sound and rigorous.

At present, MuduoLLM has not yet ventured into areas such as in‑depth mental health intervention or physical education that involves embodied interaction. The R&D team noted that amid the profound transformation of education driven by artificial intelligence, it is essential to remain committed to the fundamental values of education. It must be fully recognized that genuine emotional connection and physical exercise in real settings still rely on the personal guidance and example of teachers. MuduoLLM’s ultimate mission is not to replace teachers but to take over tedious knowledge‑based tutoring, freeing up teachers to engage more with students in the classroom and devote more attention to holistic education, including students’ mental well‑being and physical health.

Extending Benefits to Central and Western Regions: Bringing High‑Quality Resources to Every Classroom

Accurately identifying the role and direction of AI in building a leading education nation is a matter of profound importance for the country’s future and rejuvenation. Beyond easing teachers’ workload, advancing educational equity is another core mission of MuduoLLM. MuduoLLM is committed to making high‑quality teaching resources easily accessible to remote and underdeveloped areas, with low barriers to use. Through its five core application scenarios, it not only supports the professional development of frontline teachers but also helps teachers provide personalized learning companionship for students, thereby promoting the high‑quality and balanced development of basic education.

Looking ahead, the MuduoLLM R&D team has set out a clear roadmap. Currently, MuduoLLM is available free of charge to users in Beijing, with computing resources fully provided by TAL Education Group. Going forward, the team will focus on low‑cost, localized deployment and adaptation to regional teaching conditions, continuously iterating and upgrading the model.

For the central and western regions, lightweight technical solutions will be adopted to significantly reduce the hardware and software access costs for local schools. At the same time, targeted solutions will be rolled out to address specific difficulties in local teaching. In line with the principle of inclusive education, the platform will provide regular free basic services and periodic free access for public benefit, enabling teachers and students in the central and western regions to share premium AI resources with no access barriers.

"Education is not merely about the digital analysis of teaching processes, nor can teachers and students be reduced to predictable data points. The integration of AI into education must combine value‑oriented guidance with humanistic care. We always put people first and view AI as a supportive tool," said Huang Hua. Technology should always serve to cultivate people’s rational thinking, moral character, aesthetic ability, and future‑ready skills. The team remains true to its original mission of advancing technological R&D so that AI can better empower teaching and learning—a path they are steadily following.