⌘K
Change language Switch ThemeSign In
Narrow Mode
杨斌：确定 token 的中文译名，已经迫在眉睫了
腾 腾讯研究院 @腾讯研究院
One Sentence Summary
Dean Yang Bin proposes standardizing the Chinese translation of the core AI metric 'token' to 'Mo-Yuan' (模元), aiming to break cognitive barriers through precise terminology and build a bridge for industry consensus and inclusive AI in the AI era.
Summary
Written by Dean Yang Bin of Tsinghua University, this article explores the importance of 'token' as a core economic benchmark and unit of measurement in the era of AI reasoning. Drawing on the 'token factory economics' proposed at the 2026 NVIDIA GTC conference, the author points out that tokens have transcended simple technical parameters to become fundamental units possessing attributes of information, compute, and currency. The article analyzes the evolution of tokens from linguistics to AI computing units, compares the limitations of existing translations like 'Ci-Yuan' (lexical unit), 'Yu-Yuan' (linguistic unit), and 'Tuo-Ken' (transliteration), and argues for the significant advantages of 'Mo-Yuan' (model unit) in anchoring AI scenarios, aligning with Chinese measurement conventions, and accommodating future multi-modal development. The author emphasizes that a unified and accessible Chinese term is a crucial link for AI technology to move from professional circles to mass adoption and integrate into all sectors of the Chinese economy.
Main Points
* 1. Tokens have become the core economic benchmark and fundamental unit of measurement in the AI era.Tokens possess triple attributes of information, compute, and currency. As the smallest unit of AI reasoning, they directly determine the revenue capacity and core competitiveness of AI enterprises. * 2. Existing Chinese translations for 'token' have limitations and fail to capture the essence of the large AI model era.'Ci-Yuan' is limited to text, while 'Tuo-Ken' lacks semantic meaning; these translations either narrow the technical essence or increase cognitive barriers for the general public. * 3. The 'Mo-Yuan' translation offers professional rigor, industrial practicality, and mass appeal.'Mo' (Model) anchors the AI context, and 'Yuan' (Unit) represents the smallest unit. This aligns with Chinese measurement naming logic and effectively bridges the gap between professional circles and the general public. * 4. A unified Chinese translation serves as a linguistic bridge to drive AI adoption across all industries.Precise and accessible terminology lowers the barrier for traditional industries to understand AI, fostering technical consensus and the popularization of the intelligent economy.
Metadata
AI Score
83
Website mp.weixin.qq.com
Published At Today
Length 2964 words (about 12 min)
Sign in to use highlight and note-taking features for a better reading experience. Sign in now
杨斌 2026-03-18 17:00 北京
token，模元。模元，token。
杨斌 清华大学可持续社会价值研究院院长
2026年英伟达GTC大会上，黄仁勋近两小时的主题演讲，全程围绕AI推理时代的产业重构展开，堪称 AI 推理时代的产业宣言。
从硬件架构革新、数据中心转型为模元（token）工厂，到万亿级AI基建蓝图、企业经营逻辑，再到智能体与物理世界AI的未来，token这个词累计出现超过70次，成为串联整场演讲的锚点和主线。
黄仁勋以手势示意“Token King”英伟达token成本全球最低。（图片来源：GTC 2026）
针对这一定义AI时代的核心度量单位，我在2026年初提议过token的AI领域专属中文译名——模元，用以区别于区块链、网络安全等其他场景中的token译名。这次在线上越听黄仁勋的这场演讲，越让我感到这事儿有点儿急迫，“模元”作为token中文译名的推广，真的是越来越重要、越必要。
为什么？因为模元（tok en）不止是黄仁勋演讲中的高频锚点，更是思考和重构AI经济逻辑、推动AI在中国走向千行百业的关键基石；而一个精准适配、好懂易传播的中文译名——模元，正是AI从专业圈层走向全民“众技”的关键纽带。
黄仁勋在演讲中正式提出“模元（token）工厂经济学”，宣告传统数据中心已告别文件存储的旧定位，全面转型为生产模元（token）的智能工厂。固定功耗下的模元（token）每秒吞吐量、单位模元成本，直接决定AI企业的营收能力与核心竞争力，模元（token）已然成为AI时代的新大宗商品与基础度量衡。
他的这一论断，绝非技术圈层的自说自话，而是揭示了AI产业的底层逻辑：模元（token）是AI时代的核心经济标尺，兼具着信息单位、算力单位、货币单位三重属性，是AI思考的最小单元，也是算力消耗核算基准和智能服务价值度量。于此之上，全球AI产业的运行规则正在被重塑。
前一阶段公布的数据表明，全球大模型日均模元（token）消耗已达30万亿级别，中国模型调用量首次超越美国，占到全球60%以上。小到一次AI问答，大到电影级视频制作，更大到企业级模型训练，都以模元（token）计量，这使其超越一般技术参数，成为智能经济规模与活力的核心指标，也让模元的中文定名，从可以慢慢来的学术探讨变成了实践中的急迫需要。
换句话说，当一个核心且被高频使用的技术名词成为万亿级产业的核心标尺，其中文定名便不再是无关紧要的细节，而是关乎产业共识形成、技术普惠落地、公众认知普及的刚需。
因为各种可以理解的原因，当前AI行业的技术专家、企业高管与投资人，普遍直接使用英文词“token”进行交流，播客、访谈、讲座中，几乎没有统一的中文表述。这倒真不是英文水平高低的问题，即便英文基础良好，如果不怎么懂AI，也难以精准理解token的核心要义，就算翻开字典也是一头雾水；对于未接受AI专业训练、不熟悉英文技术术语的普通大众与传统行业从业者而言，生硬插入的英文名词更会带来强烈的距离感与认知隔阂。
情况是这么个情况，却并不是就该一直这么下去。我们每个人都期待AI真正融入中国千行百业、走进普通人的工作与生活，这就倒逼着我们，必须为这个核心概念，定一个通俗易懂、精严谨准、适配未来的中文译名。
从词源脉络来看，token源自古英语tāc(e)n，本义为“标志、符号、证明”，核心是“可被识别、承载特定信息或功能的基本单元”。历经中世纪商业代币、网络安全令牌、语言学“词例”的演变，进入AI大模型时代，token完成了决定性跃迁——从语言学碎片化单元，升级为AI模型可计算、可处理的最小通用单元，正式取代计算机、互联网时代的“字节”，成为AI时代的基础度量衡。
比较一下：字节是计算机物理层面的存储单位，计量机械均匀、与语义无关，仅记录数据的物理体积；而模元是智能逻辑单元，承载文本、图像、音频、视频等全模态信息，关联模型理解、推理计算、算力消耗与价值创造。这一本质区别，正是计算文明从“数据处理”走向“智能涌现”的核心标志，也让模元（token）成为贯穿黄仁勋2026GTC演讲的核心主线。
在我提出“模元”这个译法之前， AI大模型领域的token也曾有过多种中文译名，但推敲下来，发现都无法匹配AI大模型、智能体时代的核心内涵，难以打破大众的认知壁垒。比如，“词元”被“词”字锁死在文本场景，无法适配多模态、物理AI的应用形态；“语元”囿于语言范畴，窄化了token作为模型通用处理单元的本质；“义节”过度聚焦语义，忽略了token纯特征、结构化处理的属性；而“托肯”“屯”等单纯音译，徒有其音、缺乏实义，普遍接受度低，还会加重非专业人群的理解负担。这些译名要么局限于单一领域，要么缺乏度量衡的严谨性，确实无法承载token作为AI产业核心锚点的价值。
我斟酌再三提出的“模元”这个译法，是专为AI时代量身定制的意译。“模”直指大模型、多模态，锚定AI场景的核心属性；“元”代表最小基本单元，承续“字节”这类中文经典度量单位的命名逻辑，简洁直白、通俗易懂。
这一译法具备三大不可替代的优势：一是对大众友好，对中文世界的非专业受众而言，“模元”没有英文token的距离感，无需专业背景就能感知这是AI世界的基础计量单位；二是对产业实用，对产业界而言，“模元消耗量”“模元效率”“模元成本”“模元预算”等概念，能直接对应AI产业核心指标，让“模元工厂经济学”走出专业圈层，被更多人理解；三是对未来兼容，对未来发展而言，模元不局限于当下的文本推理，更适配智能体、多模态融合、物理世界AI等全场景，具备穿越技术周期的延展性。
如果试着将GTC大会上，黄仁勋演讲中的“token工厂”改为“模元工厂”，“单位token成本”改为“单位模元成本”，原本晦涩破碎的产业逻辑表达会立刻变得顺畅且清晰易懂：从业者明白模元是AI生产的基础原料；投资者理解模元效率是企业降本增效的关键；普通大众懂得，“模元”就是AI时代的“字节”，是走进智能经济的入门钥匙。技术的真正价值，在于从专业重器走向全民众技，而朗朗上口、易于传播的中文译名，正是AI普及的必要一步。
我反复阐释“模元”这个译法的初衷，并不是纠结较真、咬文嚼字，有点儿执意想请大家把“模元”作为惯用词统一起来，真是想为中国AI高普及率搭建必要的语言桥梁，打破专业与大众之间的认知壁垒，让AI核心词汇更易读易懂，让传统行业出身的人在认知和把握AI本质的路上，不用先越过什么语言门槛就能径直探索AI时代的基本规律，促进AI这个进行时很强的科技在中国走向千行百业、走向全民普惠。
按照荀子有言“名无固宜，约之以命”，所谓约定俗成，再合理的“名”，没人用、形不成共识也是白费。期待大家在学术研究、产业分析、政策讨论与媒体报道中，有意识地采纳并使用“模元”这一译法，共同推动这个有根（词源清晰）、有义（内涵准确）、有未来（延伸性强）的词汇，成为理解、参与和拥抱AI时代的日常用语。
马年龙抬头将至，AI 时代的模元经济，正加速演进大变局，却也还只是开局。期待不论产业还是社会，专业抑或日常，以“模元”为计量，向智能而涌现。!Image 10 推荐阅读
腾讯研究院：《春节AI热潮之后，网民真的开始用AI了吗？｜T-ask调研》
!Image 11 👇 点个“在看”分享洞见 阅读原文 跳转微信打开
腾 腾讯研究院 @腾讯研究院
One Sentence Summary
Dean Yang Bin proposes standardizing the Chinese translation of the core AI metric 'token' to 'Mo-Yuan' (模元), aiming to break cognitive barriers through precise terminology and build a bridge for industry consensus and inclusive AI in the AI era.
Summary
Written by Dean Yang Bin of Tsinghua University, this article explores the importance of 'token' as a core economic benchmark and unit of measurement in the era of AI reasoning. Drawing on the 'token factory economics' proposed at the 2026 NVIDIA GTC conference, the author points out that tokens have transcended simple technical parameters to become fundamental units possessing attributes of information, compute, and currency. The article analyzes the evolution of tokens from linguistics to AI computing units, compares the limitations of existing translations like 'Ci-Yuan' (lexical unit), 'Yu-Yuan' (linguistic unit), and 'Tuo-Ken' (transliteration), and argues for the significant advantages of 'Mo-Yuan' (model unit) in anchoring AI scenarios, aligning with Chinese measurement conventions, and accommodating future multi-modal development. The author emphasizes that a unified and accessible Chinese term is a crucial link for AI technology to move from professional circles to mass adoption and integrate into all sectors of the Chinese economy.
Main Points
* 1. Tokens have become the core economic benchmark and fundamental unit of measurement in the AI era.
Tokens possess triple attributes of information, compute, and currency. As the smallest unit of AI reasoning, they directly determine the revenue capacity and core competitiveness of AI enterprises.
* 2. Existing Chinese translations for 'token' have limitations and fail to capture the essence of the large AI model era.
'Ci-Yuan' is limited to text, while 'Tuo-Ken' lacks semantic meaning; these translations either narrow the technical essence or increase cognitive barriers for the general public.
* 3. The 'Mo-Yuan' translation offers professional rigor, industrial practicality, and mass appeal.
'Mo' (Model) anchors the AI context, and 'Yuan' (Unit) represents the smallest unit. This aligns with Chinese measurement naming logic and effectively bridges the gap between professional circles and the general public.
* 4. A unified Chinese translation serves as a linguistic bridge to drive AI adoption across all industries.
Precise and accessible terminology lowers the barrier for traditional industries to understand AI, fostering technical consensus and the popularization of the intelligent economy.
Key Quotes
* Mo-Yuan is an intelligent logic unit that carries multi-modal information such as text, images, audio, and video, linking model understanding, inference computation, compute consumption, and value creation. * The true value of technology lies in its transition from a professional tool to a skill for the masses, and a catchy, easy-to-spread Chinese name is a necessary step for AI popularization. * Mo-Yuan (token) has become the new commodity and fundamental unit of measurement for the AI era. * When a core, frequently used technical term becomes the primary benchmark for a trillion-dollar industry, its Chinese naming is no longer a trivial detail. * We look forward to a future where, whether in industry or society, professional or daily life, we use 'Mo-Yuan' as the standard measure to emerge toward intelligence.
AI Score
83
Website mp.weixin.qq.com
Published At Today
Length 2964 words (about 12 min)
Tags
Token
Mo-Yuan
AI Economics
Terminology Translation
Large Models
Related Articles
* Yao Shunyu Lectures Face-to-Face with Tang Jie, Yang Zhilin, and Lin Junyang! Four Schema Heroes Debate Heroes at Zhongguancun * 131. Yin Qi's First Interview as Chairman of StepStar: The Temptation of Smart People, the Brutal Elimination Race, Bets, and Hyper-Multivariate Equations * AI Health Assistants: A Rising Global Tide * After Large Models: Rewriting the Division of Labor Between Humans and Machines | A 10,000-Word Roundtable Transcript * Zhang Xiaoyu: Why I Became a Firm AI "Adventist"? * AI Will Bring an Economic Explosion, but the Fuse is Long | Hao Paper Talk * What AGI Narratives Are AI Giants in China and the US Describing? * Vol.94 | A Conversation with Zibianliang Robotics: Three Bold Assertions on Robot Data * Tech Enthusiast Weekly (Issue 381): What Are China's AI Large Model Leaders Thinking? * Faith and Breakthroughs: 2026 Artificial Intelligence Trend Outlook HomeArticlesPodcastsVideosTweets