Beijing Rolls Out China’s First Token Economy Action Plan
Beijing has issued China’s first province-level special policy dedicated to the “token economy,” a plan that seeks to build a complete industrial chain around the tokens that power large artificial intelligence models. The action plan, jointly issued by the Beijing Municipal Bureau of Economy and Information Technology and the Beijing Municipal Development and Reform Commission on September 15 and publicly released on September 18, positions the token as a measurable, tradeable and billable unit of the AI economy.
The “Beijing Action Plan for Accelerating the Development of the Token Economy 2026-2028” calls for strengthening production, quality evaluation, distribution and application of tokens (ciyuan), the Chinese official translation of the term Token, which was designated by the National Data Administration in March 2026. The policy takes effect from publication until December 31, 2028.
What Is the Token Economy?
A token is the smallest unit of measurement that large AI models use to process information, including text, code, images, audio and video. As AI moves deeper into production and daily life, tokens have become a key link connecting computing power, models and application scenarios, and are increasingly serving as the unit of measurement, settlement and statistics for AI services.
“A token can be understood as the basic measurement unit for AI processing text, code and other information,” said Zhang Jinrui, director of the Digital Industry Division at the Beijing Municipal Bureau of Economy and Information Technology, as reported by Xinhua News. “Every time a large model answers a question, generates a piece of code or calls an agent service, it involves the processing and generation of tokens.”
The policy implements the State Council’s AI Plus initiative and the 2026 Government Work Report’s first-ever call to build new forms of the intelligent economy. It reflects a broader shift in focus: where the market once concentrated on how much computing power and how many models existed, attention is now turning to how often models are called, how efficiently, and what intelligent services those calls ultimately produce.
Ten Measures Across Six Areas
The action plan lays out ten measures spanning token factory construction, core technology, quality evaluation, distribution platforms, agent cultivation, the agent software ecosystem, overseas service capacity, data ontology technology and token engineer training.
On the production side, Beijing will build world-class “token factories” and establish a tiered evaluation standard covering model adaptation count, token throughput speed, first-token latency, cache hit rate and energy efficiency (PUE), according to Xinhua via Qstheory. “Token factories have the attribute of platform distribution and are at the forefront of external applications,” Zhang said.
To address quality, the plan calls for a token quality evaluation system and support for third-party testing platforms, with objective performance and quality standards intended to inform pricing and routing. “People can easily see token consumption volume, but must further pay attention to the level of intelligence behind the tokens,” said Chen Guangming, deputy director of the Digital Industry Division, noting that token factory standards and evaluation systems remain inconsistent and that “token idling” persists.
The plan also envisions token distribution platforms with cross-model, cross-modal and cross-region access, offering unified interfaces, model aggregation and token metering and settlement. Such platforms would be included in the layout of new-type infrastructure. “Token prices are mainly formed through full market competition; the government’s more important role is to improve the basic conditions needed for market operation, such as token quality evaluation, metering and distribution,” Chen said.
On applications, Beijing encourages platform companies and agent service providers to scale up agents in information consulting, commercial services and daily life services, while opening agent application scenarios in industrial manufacturing, scientific research, biomedicine and education. The policy also breaks new ground by proposing, for the first time, exploration of a “data ontology” technical paradigm to convert industry knowledge, process parameters and equipment data into data products usable by large models. It additionally calls for “token engineer” reskilling programs covering model tuning, inference acceleration, platform operations and security compliance.
Beijing’s Advantages and Regional Competition
Beijing is positioning the token economy as a core advantage. In the first half of 2026, the city added 22,000 P of intelligent computing power, reaching a total supply of 82,000 P, backed by the Galaxy Compute Corridor project and a nationwide computing interconnection network. More than half of domestic token production is concentrated in Beijing, and ByteDance’s Doubao model alone exceeds 180 trillion daily token calls, according to Zhang.
Separately, the Beijing E-Town (Yizhuang) development zone unveiled its own token economy layout in late June. Its “Beijing No.1 Token Factory” reached a phase-one daily capacity of more than 1.4 trillion tokens, with half of tasks completed within six seconds and 90 percent within ten seconds, as the zone’s People’s Daily Beijing report described. Yizhuang aims to build four 10,000-card-class token factories and reach a 400 billion yuan regional intelligent-economy scale by 2030.
Beijing is not alone. Multiple localities, including Guangdong, Sichuan, Hainan, Jiangsu, Hubei and Guizhou, are moving on the token economy, according to Economic Reference News. Tools include token coupons: Sichuan offers subsidies of up to 20 percent of actual usage, capped at 1 million yuan, while Yangzhou supports up to 500,000 yuan per applicant per year and Wuhan’s Jiang’an district plans 10 million yuan annually. Guizhou is emphasizing intelligent computing, targeting 240 EFLOPS by 2028, while Guangdong is focusing on cross-border token services.
Challenges and Outlook
Experts caution that the token economy remains in an exploratory stage, with unresolved issues around standards. “Token standards and metering systems are not yet unified,” said Chen Duan, director of the Digital Economy Integration Innovation Center at the Central University of Finance and Economics. “Cross-province, cross-platform mutual recognition remains difficult.”
Wang Chunhui, a professor at Nanjing University of Posts and Telecommunications, said regions mastering core underlying technologies and high-quality data may occupy the top of the industrial chain, while those with rich application scenarios will become the main battlefield for converting token value. He also cautioned that token consumption is currently recorded simply as a cost, which makes it hard to reflect efficiency gains, and called for a “token input-business output” accounting system.
Beijing frames its approach as one that does not simply pursue token consumption volume. “The token economy is still in the exploratory stage. The market is here, and the demand is here. Beijing will become the experimental field for the token industry,” Zhang said, inviting businesses to pilot related operations in the city.
For now, Beijing frames the token economy as a long-term experiment rather than a quick-win spending push. As the policy runs through 2028, the coming months will test whether the city can turn its concentration of models, platforms and talent into a functioning market for tokens as a tradeable unit of the AI economy.