Tsinghua Professor Proposes "模元" as Official Chinese Name for AI's Core Unit 'Token'

Beijing, March 20, 2026 – As 'Token' becomes a defining metric of the AI era, a leading Chinese academic has put forward an official Chinese translation to standardize its use and public understanding.

Professor Yang Bin, Dean of the Institute for Sustainable Social Value at Tsinghua University, published a proposal through the Tencent Research Institute, advocating for "模元" (pronounced "mó yuán") to be adopted as the exclusive Chinese name for 'Token' in the context of artificial intelligence.

The proposal comes as the term 'Token' gains unprecedented prominence. At NVIDIA's GTC 2026 conference, CEO Jensen Huang reportedly mentioned it over 70 times, underscoring its central role. However, a lack of a consistent, intuitive Chinese translation has created a barrier to public comprehension and industry discourse.

The Case for "模元"

Professor Yang's analysis critiques existing translation attempts:

• "词元" (Cí Yuán - Word Unit): Too narrowly tied to text, failing to adapt to multimodal and physical AI applications.
• "语元" (Yǔ Yuán - Language Unit): Confined to the linguistic domain, which undersells Token's role as a universal model processing unit.
• "义节" (Yì Jié - Semantic Segment): Over-emphasizes meaning, neglecting Token's nature as a pure, structured feature.
• "托肯" or "屯" (Tuō Kěn / Tún): Mere phonetic transliterations that lack semantic value and increase the learning burden for non-experts.

"模元" is designed as a purpose-built, meaningful translation for the AI age. The character "模" (mó) directly points to large models and multimodality, anchoring the term in the core AI context. The character "元" (yuán) signifies the smallest fundamental unit, continuing the naming logic of classic Chinese measurement terms like "字节" (byte).

Three Key Advantages

Professor Yang argues that "模元" offers three irreplaceable advantages:

1. Public Accessibility: For the general Chinese-speaking public, "模元" removes the distance of the English term 'Token.' It intuitively conveys the concept of a basic AI measurement unit without requiring a technical background.
2. Industrial Utility: For the industry, terms like "模元消耗量" (Token consumption), "模元效率" (Token efficiency), and "模元成本" (Token cost) can directly correspond to core AI business metrics. This would help demystify "Tokenomics" and make it accessible beyond specialist circles.
3. Future Compatibility: The term is not limited to current text-based inference. It is designed to be adaptable across future developments, including AI agents, multimodal fusion, and physical-world AI, ensuring its relevance across technology cycles.

The proposal has ignited discussion within China's tech community about the importance of precise, accessible terminology as foundational AI concepts enter the mainstream.