At the 2026 World Computing Conference, National Data Administration head Liu Liehong said tokens have become the value anchor linking compute, models and data, with a token production-circulation-consumption ecosystem taking shape. The agency will expand token service scenarios, build metering, billing and quality assessment standards, and explicitly curb low-quality redundant construction.
Token
The minimal unit of text processed by large models, used as a metering and delivery unit linking compute, models and data.
Curated Intelligence
The Ramp AI Index, built on transaction data from over 85,000 US companies, shows AI token volume hitting a record high while per-token prices fell 21.2 percent in six weeks, with API spending down 6.3 percent from its peak two weeks earlier. Rising usage paired with falling unit prices means the industry still needs extraordinary growth to recoup its planned capacity buildout.
Lenovo has unveiled its Wanquan Hybrid Token Factory brand in Zhangjiakou, Hebei, planning a 15,000P super-node cluster that it claims will produce 2 trillion tokens annually at 70% lower cost. The "green power in, tokens out" model aims to industrialize token production by pooling heterogeneous compute and models, addressing enterprises' soaring token bills and idle capacity. However, details on scheduling architecture, model lineup and pricing remain undisclosed.
The China Science Writers Association released its 2026 top 10 science popularization terms, with AI-related entries such as "AI+", "agent", "token" and "embodied intelligence" taking seven of the ten slots, reflecting rising public demand to understand AI as it spreads into daily life. The list is a year-end roundup rather than a technical release and carries no verifiable new data.
Bloomberg reports that Apple is significantly cutting internal budgets for employee AI usage to curb rising costs. Daily token consumption caps for some engineers have dropped from $300 to $200, with plans to reduce them further to $100 or lower. This move reflects a broader industry shift from encouraging large-scale experimentation to rigorous cost-benefit analysis, mirroring similar actions by Microsoft, Tesla, and others.
Senior developer Nolan Lawson benchmarked 12 JavaScript frameworks using Claude Opus 5.5 for AI Agent coding tasks. Results showed that building the same app with Angular cost nearly double ($4.97 median) and consumed almost twice the tokens (9.76M) compared to the cheapest option, Hono+htmx, prompting a reevaluation of frontend tech stack choices in the AI era.
OpenAI announced that starting in December, it will merge ChatGPT's chat interface with Codex's agentic work features, implementing a unified token budget across Plus, Pro, and Business plans. This shift signals the end of unlimited prompting models, requiring users to manage usage limits based on remaining credits for complex tasks.
OpenAI announced that starting in December, usage quotas for ChatGPT and Codex will be merged into a shared pool for Plus, Pro, and Business plans. The key change is the shift from message-based counting to unified token-based billing, with only Instant chat exempted. This move aims to integrate agentic capabilities and tighten previous frequent quota reset policies.
OpenAI announced that starting in December, usage allowances for Plus, Pro, and Business plans will be shared across ChatGPT and Codex. The new system measures usage in tokens, with advanced models and complex tasks consuming more quota, aiming to unify agentic capabilities and simplify billing.
The article argues that developers should build their own LLM proxies instead of relying solely on frontend interfaces to gain underlying visibility into inference processes, actual token consumption, and data flow. The author contends that official usage statistics are often misleading, and intercepting requests via a middleware layer enables more precise cost control and security auditing.