What is 词元?
CIYUAN = 词元 = Token
The official Chinese name for "token" — announced by China's National Data Bureau at the 2026 China Development Forum.
中国国家数据局在中国发展高层论坛2026年年会上正式公布的"token"官方中文名称。
140万亿/天
Daily Token Usage (March 2026) / 日均词元调用量
1,000×
Growth in 2 Years / 两年增长倍数
Mar 23, 2026
Official Announcement / 官方发布日期
Definition / 定义
Three Terms, 同一概念
CIYUAN, 词元, and "token" all refer to the same concept — 词元、CIYUAN 与"token"三者指向同一概念 the smallest information unit that large AI models process. 即人工智能大模型处理信息的最小信息单元。
English
英文
token
The fundamental unit of data that large language models process. For example: "I love China!" might split into four tokens: "I", "love", "China", "!"
Romanized
拼音
CIYUAN
The official romanized designation. Proposed by Professor Qiu Xipeng (Fudan University, 2021) and formally adopted by China's National Data Bureau in March 2026.
中文
Chinese
词元
"词"覆盖字和词的范围，"元"是最小基础单元。两个汉字合在一起，精准描述了 token 在大模型中扮演的角色。
Official Source / 官方来源
On March 23, 2026, at the 2026 China Development Forum annual conference, Liu Liehong (刘烈宏), Director of the National Data Bureau (国家数据局), officially introduced 词元 (CIYUAN) as the standardized Chinese name for "token". The term was subsequently reported by People's Daily and widely circulated in official Chinese media.
2026年3月23日，中国国家数据局局长刘烈宏在中国发展高层论坛2026年年会上正式将"token"的中文名称定为"词元"（CIYUAN）。
By the Numbers / 数据说话
A 1,000× Surge in Two Years / 两年增长超千倍
China's daily token consumption tells the real story of the AI boom — 中国日均词元消耗量的变化，折射出 AI 产业的真实面貌 not in benchmark scores, but in industrial throughput. 不是评测分数，而是工业产能。
Early 2024
100亿/天
Daily token usage in China at the start of 2024.
2024年初，中国日均词元调用量为1000亿。
End of 2025
100万亿/天
A 400× increase in roughly 1.5 years — as reported by the National Data Bureau.
国家数据局披露，一年多增长400多倍。
March 2026
140万亿/天
140 trillion tokens per day — over 1,000× growth from the start of 2024.
突破140万亿，较2024年初增长超千倍。
“Token（词元）不仅是智能时代的价值锚点，更是连接技术供给与商业需求的‘结算单位’，为商业模式的落地提供了可量化的可能。”
“Token [CIYUAN] is not only the value anchor of the intelligent era, but also the ‘settlement unit’ connecting technology supply and commercial demand, providing quantifiable possibilities for business model implementation.”
Why It Matters / 为何重要
A Quiet Shift in AI's 叙事权的悄然切换
"词元" is not just a good translation — "词元"不只是一个好翻译 it is a signal that China's AI narrative has completed an identity switch: from "we are also catching up" to "we are exporting production capacity." 它是中国 AI 叙事完成身份切换的信号：从"我们也在追赶"，变成"我们正在输出产能"。
Benchmark Rankings
/ 评测分数时代
- Which model scores higher on MMLU, HumanEval, GPQA?
- English-language benchmarks as the universal standard
- Every Chinese model launch measured against GPT-4o
- The ruler is made by others — you just compete on it
谁的 benchmark 更高？参数量更大？评测分数更领先？尺子是别人造的。
Token Volume
/ 词元产能时代
- How many tokens consumed per day? How many API calls?
- Industrial capacity as the metric — a domain China knows well
- China's weekly token usage: 4.12 trillion vs US: 2.94 trillion (People's Daily, 2026)
- The ruler is built at home — the standard is set by usage scale
词元消耗量、调用量曲线，尺子是中国造的，用规模来定义话语权。
Expert Definition / 专家定义
“A token is the discrete unit for data processing in natural language algorithms. With the rise of large models, tokens provide a unified representation for diverse modalities — enabling cross-modal understanding and generation. From text subwords to visual patches, tokenization enhances data processing efficiency.”
NVIDIA & Jensen Huang / 英伟达 & 黄仁勋
At NVIDIA's GTC 2026 conference, CEO Jensen Huang explicitly stated that the token is the foundational building block of the new AI era. The English term is "token"; the Chinese term is 词元. Both sides are crowning the same concept — in different languages, with equal weight.
英伟达CEO黄仁勋在2026年GTC大会上明确指出，token 是新 AI 时代的基础构建单元。英文叫 token，中文叫词元，两边同时在给这个概念加冕。
FAQ / 常见问答