The Transition of Token: From Floor Price to Rational Tiered Pricing
English Version
At midnight on August 17, 2026, DeepSeek officially launched its peak-valley pricing mechanism. The cached hit input cost of its V4 Pro model surged from 0.025 RMB to 0.3 RMB per million tokens, a staggering increase of 1100%. On the exact same day, Unexhub officially launched a comprehensive financial clearing and settlement system, building a standardized computing power trading center focused on token aggregation and distribution. It features core advantages including instant settlement, zero commission for early suppliers, and dynamic API bidding mechanisms.
The synchronous release of the two major updates marks a pivotal industry shift: the token market is stepping out of the single-dimensional "floor price" competition and entering a new era of hierarchical and rational pricing.
General Token Deflation, Enterprise AI Budget Inflation
There is a counter-intuitive industry trend worth analyzing. Taking OpenAI’s pricing as the benchmark, GPT-4 was priced at $30 for input and $60 for output per million tokens when launched in March 2023. Today, a large number of mainstream inference models have driven the general token price down to mere cents, with some ordinary token prices plummeting by up to 99% in three years.
While universal token costs continue to hit new lows, enterprises’ overall AI operational expenditures have not decreased at all. The core reason lies in scenario upgrading: enterprises are shifting from simple Q&A interactions to complex workflows such as autonomous Agent collaboration and intelligent code generation. These high-value tasks consume 10 to 100 times more tokens than basic conversations.
A typical case is Uber: after opening AI programming tools to 5,000 engineers, the company exhausted its full-year 2026 AI budget within just four months. This is not an isolated incident. The truth of the current market is clear: ordinary general-purpose tokens are approaching zero-cost deflation, high-end capable tokens are forming tiered value premiums, and enterprise AI costs are expanding alongside upgraded calling structures.
Cheap floor-price tokens are ubiquitous, but tokens capable of supporting complex industrial tasks have never been low-cost.
Tiered Pricing: The Inevitable Upgrade from Floor Price to Rational Price
A June industry report released by Changjiang Securities pointed out that the token market is evolving from a unified flat market to a refined tiered market. Economy and neutral lightweight models are depreciating rapidly with half-lives of 1.10 years and 1.55 years respectively, faster than traditional Moore’s Law iteration. In contrast, flagship high-performance models no longer follow the universal price reduction trend and maintain stable value pricing.
This is not a commercial monopoly strategy of model vendors, but an inevitable result of industrial maturity.
Representative tiered pricing cases are widespread in the industry. Zhipu GLM-5.1, tailored for programming and Agent complex scenarios, is priced at 6–8 RMB for input and 24–28 RMB for output per million tokens. Its lightweight version GLM-4.7, built for daily casual tasks, only charges 2–4 RMB for input and 8–16 RMB for output, creating a three-fold price gap for differentiated scenarios under the same technical system.
Major cloud vendors have also fully rolled out tiered billing. Alibaba Cloud launches three independent billing systems for different scenarios: real-time pay-as-you-go, token monthly packages, and exclusive coding subscriptions. In May, ByteDance Doubao launched a commercial paid system, implementing separate pricing for high-computing scenarios including PPT generation and data analysis.
Shao Yilei, Dean of the School of AI and Finance at East China Normal University, commented: "Vendor tiered token pricing and enterprise dual-standard token accounting are the two core symbols of mature large-model commercialization."
Unexhub: Elevating Tiered Pricing to Dynamic Market Trading
DeepSeek realizes time-based tiered pricing — the same token has different prices for peak and off-peak hours. Unexhub further innovates the industry by turning tokens into freely tradable market assets, just like securities.
The core mechanisms of Unexhub’s token trading ecosystem are as follows:
Token Aggregation & Unified Distribution: Gathers scattered computing power suppliers of all scales to form a unified open trading market. Small and medium-sized suppliers no longer need to build independent platforms or develop customer resources, enabling one-click token sales and rapid market access.
Instant Financial Settlement: Completely solves the traditional industry pain points of long account periods, complicated reconciliation and fund occupation. Real-time fund recovery eliminates financial pressure for small and medium-sized resource suppliers.
Zero Commission for Early Settlers: Early cooperative suppliers enjoy zero platform service fees, accessing mature high-traffic sales channels at zero cost and retaining full profit margins.
Dynamic API Bidding Mechanism: Token prices float autonomously based on real-time market supply and demand. Prices rise during peak demand to increase supplier revenue and drop during oversupplied off-peak periods to deliver ultra-low costs for developers.
Essentially, tokens are no longer fixed-price industrial consumables, but financial commodities driven entirely by market supply and demand.
The infrastructure for token trading is being comprehensively deployed nationwide. In March 2026, China’s daily token calling volume exceeded 140 trillion times, a more than 1000-fold increase compared with early 2024. In late May, Inner Mongolia Token Trading Platform, the country’s first full-stack green computing AI trading platform, launched one-stop services including heterogeneous computing aggregation, high-quality model access and token settlement. In July, the Yangtze River Delta (Jiaxing) Token Operation Center was officially inaugurated, gathering core mobile computing resources to lower the threshold for enterprise intelligent transformation via resource aggregation and intelligent scheduling.
Industry institutions have reached a consistent judgment: tokens will become a new type of bulk commodity. After market maturity, tokens will form a complete tiered pricing system based on response speed and intelligent capabilities, ranging from free basic tiers to ultra-high-speed premium tiers (nearly $150 per million tokens), opening up huge commercial space for high-end reasoning scenarios. JPMorgan predicts that the average token price will drop by more than 40% in the next 12 months, with ordinary lightweight models fully commoditized.
Enterprise Cost Management: Survival Strategies for the Tiered Token Era
The rise of tiered pricing and token trading markets has completely reshaped enterprise AI cost management logic.
First, computing power costs have evolved from fixed rigid expenditures to controllable tradable assets. Instead of the simple binary choice of "use or not use", enterprises can precisely control costs from multiple dimensions: usage time, model grade and purchasing channel. Scenario-based precise cost-benefit calculation has become a necessary capability for enterprise AI operation.
Second, dynamic model routing has become the core of cost optimization. Mature AI applications no longer rely on a single fixed model. Instead, they match lightweight models for simple Q&A, flagship models for complex reasoning, and schedule non-urgent batch tasks to off-peak low-price periods to maximize cost performance.
Third, token trading turns passive consumption into active value operation. Enterprises can stock up on low-cost tokens during off-peak periods in anticipation of peak demand; resource suppliers can sell surplus computing power at premium prices during peak hours. Tokens have completely completed the transformation from disposable consumables to value-preservable tradable assets.
DeepSeek realizes time-based stratification, while Unexhub realizes supply-demand-based market segmentation. Together, they build the complete economic logic of the token industry: the era of single floor-price competition is over, and rational tiered pricing has become the new industry standard.
This is not a simple price increase, but the maturity of the AI pricing system. General tokens face continuous deflation, while high-value scenario-based tokens enjoy value premiums. A complete token price spectrum covering peak-valley differences, market bidding and hierarchical differentiation has taken shape.
For engineers and practitioners, the working logic has also upgraded: it is no longer about choosing the cheapest model for all scenarios, but matching the most suitable model at the most reasonable time and cost.
The transition from token floor price to rational price proves that the AI industry is moving from blind low-price involution to refined, accountable and sustainable commercial operation.
As a global token trading infrastructure leader, Unexhub rebuilds the token industry’s operational logic and builds a one-stop computing power trading center for token aggregation and distribution. The platform integrates full-scale AI resources, covering top mainstream models such as Gemini, ChatGPT, Claude, DeepSeek and GLM, as well as high-quality customized models from individual developers and SMEs.
Driven by multi-supplier public bidding and millisecond-level intelligent routing, the platform delivers extreme cost advantages for end users. Developers pay based on actual token consumption, with the lowest discount reaching 30% of the official market price, perfectly supporting daily development and high-intensity commercial deployment. For suppliers, one-click access and real-time sharing settlement greatly lower the threshold for model commercialization, helping all participants capture the dividends of the booming token economy.
Token 价格跃迁:从地板价内卷,走向分层理性定价
中文版本
2026 年 8 月 17 日零点,DeepSeek 正式落地峰谷定价机制,V4 Pro 模型缓存命中输入单价从 0.025 元 / 百万 Token 飙升至 0.3 元 / 百万 Token,涨幅高达 1100%。同日,Unexhub 平台正式上线全套金融清结算体系,打造专注 Token 聚合分发的标准化算力交易中心,落地即时结算、早期供应商零佣金、API 动态竞价三大核心能力。
两件行业关键事件同日落地,印证同一核心趋势:Token 市场彻底告别单一粗暴的「地板价」内卷,全面迈入分层定价、价值匹配的理性新时代。
通用 Token 持续通缩,企业 AI 账单反而通胀
行业存在一个反直觉但真实的现状:Token 越来越便宜,企业 AI 成本却越来越高。
以行业标杆定价为参照,2023 年 3 月 GPT-4 首发定价为输入 30 美元、输出 60 美元 / 百万 Token。时至今日,大量入门级、通用推理模型将 Token 价格打到极致低位,部分通用算力单价三年内最大降幅达 99%,近乎免费。
但海量企业的真实 AI 运营账单并未下降。核心原因在于应用场景全面升级:企业早已跳出基础问答场景,大规模落地智能 Agent 协作、全自动代码生成、复杂业务推理等高阶工作流。这类场景的 Token 消耗量,是普通对话场景的十倍甚至百倍。
典型案例极具说服力:Uber 为 5000 名工程师开放 AI 编程工具后,仅四个月就耗尽了 2026 全年 AI 预算。这并非个例,而是行业共性:通用基础 Token 持续探底通缩,高阶能力 Token 形成价值分层,企业调用结构升级,直接带动整体算力支出通胀。
行业现状已然清晰:廉价地板价 Token 随处可得,但能够支撑复杂工业级任务、稳定落地高阶业务的优质 Token,从未真正低价。
分层定价:Token 从低价内卷到理性价值的必然升级
长江证券 6 月行业报告明确指出:Token 市场已从统一平价市场,迭代为精细化分层市场。经济型、轻量通用模型降价半衰期仅 1.10 年、1.55 年,迭代降价速度超越传统摩尔定律;而旗舰高端模型彻底脱离低价赛道,坚守价值定价,不再盲目降价。
这并非厂商垄断套路,而是行业商业化成熟的必然结果。
目前全行业已全面普及分层定价:智谱 GLM-5.1 主打编程、Agent 复杂场景,定价为输入 6–8 元、输出 24–28 元 / 百万 Token;轻量化的 GLM-4.7 适配日常轻量问答,定价仅输入 2–4 元、输出 8–16 元,同品牌、同算力体系下,场景价差高达三倍。
头部云厂商同步跟进分层计费:阿里云推出实时按量、Token 包月套餐、编程专属订阅三套独立计费体系;5 月字节豆包上线商用付费体系,对 PPT 生成、大数据分析等高算力消耗场景单独定价。
华东师范大学上海人工智能金融学院院长邵怡蕾总结:「厂商分层定价、企业双层记账,两套体系的成熟落地,才是大模型行业真正走向商业化的标志。」
Unexhub:将静态分层定价,升级为动态市场化交易
DeepSeek 实现了时间维度的分层定价—— 同一 Token,分高峰、低谷差异化计价;而 Unexhub 完成了行业更进一步的革新:打破固定定价模式,让 Token 像证券一样自由竞价、市场化交易。
平台核心运营模式清晰且极具行业颠覆性:
全域 Token 聚合分发:整合市场内所有量级的算力供应商资源,搭建统一公开交易渠道。中小算力服务商无需自建平台、无需自主拓客,一键入驻即可批量出货、高效变现。
资金即时清结算:彻底解决传统算力行业账期漫长、对账繁琐、资金积压的痛点,调用收益实时回收,将中小供应商的资金周转压力降至最低。
早期入驻零佣金:平台对早期合作供应商免收服务费,零成本开放成熟流量与交易体系,服务商独享全部收益,最大化入驻红利。
API 动态竞价机制:Token 价格随市场供需实时浮动,高峰需求旺盛时价格上行,提升供应商收益;低谷资源过剩时价格下探,让利开发者,形成良性市场平衡。
这套模式的核心变革,是彻底重构 Token 属性:Token 不再是一成不变的固定价格消耗品,而是由市场供需自主定价的可交易金融资产。
与此同时,国内 Token 交易基础设施正在全域落地、快速完善。2026 年 3 月,全国日均 Token 调用量突破 140 万亿次,较 2024 年初暴涨超 1000 倍;5 月底,全国首个绿色算力全栈 AI 交易平台 —— 内蒙古词元交易中心正式上线,实现异构算力聚合、优质模型接入、Token 交易结算一站式闭环;7 月,长三角(嘉兴)Token 运营中心揭牌落地,整合移动核心算力,通过资源聚合与智能调度,大幅降低企业智能化转型门槛。
行业预判已然明确:Token 将成为数字经济时代全新大宗商品。市场成熟后,行业将根据响应速度、智能等级形成完整定价梯队,从免费基础版到超高速高端版(约 150 美元 / 百万 Token),全面打开高阶推理场景的商业化空间。摩根大通预测,未来 12 个月 Token 均价将再降 40% 以上,普通轻量化模型彻底进入商品化时代。
企业算力成本管理:新时代的核心生存能力
分层定价 + 市场化交易的普及,彻底改写了企业 AI 成本管理逻辑。
第一,算力成本从固定开支变为可调控资产。企业不再只有「用或不用」两个选择,可从使用时段、模型档位、采购渠道多维度精准控本,根据业务场景精细化核算 Token 投入性价比。
第二,模型动态路由成为核心降本手段。成熟的 AI 应用不再单一依赖某一款模型,而是智能匹配调度:简单问答用轻量低价模型,复杂推理用旗舰高精度模型,非紧急批量任务错峰低谷执行,兼顾效果与成本。
第三,Token 从被动消耗变为主动经营的期货资产。企业可预判业务峰值,在低谷低价时段囤储算力;服务商可在需求高峰溢价出货。Token 彻底摆脱消耗品属性,具备了保值、交易、增值的金融属性。
DeepSeek 定义了「时间分层」,Unexhub 定义了「供需分层」,二者叠加,构筑了完整的 Token 经济新范式:行业彻底告别无序低价内卷,进入价值匹配的理性定价时代。
这不是涨价,而是行业定价体系的成熟。通用基础 Token 持续通缩、高阶场景 Token 价值溢价,峰谷差异、供需竞价、场景分层交织,形成完整的 Token 价格光谱。
对开发者与企业而言,运营思维彻底升级:不再是无脑选用最低价模型,而是在正确的时间、用正确的模型、花合理的成本落地业务。
Token 从地板价到理性价的跃迁,不是 AI 变贵了,而是 AI 行业终于告别粗放式发展,迈入精细化算账、价值化竞争的成熟阶段。
在全新的 Token 市场化浪潮下,Unexhub 重构全球 Token 交易逻辑,搭建一站式 Token 聚合分发算力交易中心。平台全面入驻 Gemini、ChatGPT、Claude、DeepSeek、GLM 等头部大模型厂商,同时吸纳个人开发者、中小企业的优质定制化 AI 资源,依托海量供应商竞价机制与毫秒级智能路由,为用户打造极致低价的调用体验。
平台核心权益极具优势:开发者按实际 Token 消耗量付费,最低可享市场官方价 3 折,高强度商用调用也能极致控本;供应商可一键入驻平台,调用分成实时结算、高效变现,零门槛拥抱 AI 算力经济红利。