¥19 Monthly Subscription vs. 12% Commission: AI Agents Are Testing Two Completely Different Ways to Make Money
Bilingual Edition | English First, Chinese Below
In August 2026, the commercialization of AI Agents in China took two very different paths on the same day.
Alibaba’s Qwen launched paid memberships for its AI Office Assistant, with monthly plans ranging from ¥19 to ¥128 across three tiers. At the same time, ByteDance’s Doubao began charging a 12% channel service fee on hotel orders completed after users entered the transaction flow through Doubao and were redirected to Douyin Life Services. The combined rate consists of an 11.4% software service fee and a 0.6% payment fee.
One model sells capability. The other monetizes transactions.
Both point toward the same fundamental question:
How should AI Agents actually make money?
Qwen’s Logic: Subscription Revenue for “Work on Your Behalf”
Qwen’s AI Office Assistant uses a three-tier membership structure:
Advanced: ¥19/month
Elite: ¥49/month
Flagship: ¥128/month
Annual plans are priced at approximately ¥200, ¥568, and ¥1,499 respectively, with usage quotas increasing by roughly 2×, 5×, and 20× across the tiers.
So what does ¥19 actually buy?
The three membership tiers provide access to Qwen’s flagship model and capabilities such as local and cloud-based workflows, remote smartphone control, Skills, Connectors, and the ability to generate deliverables including webpages, PowerPoint presentations, Word documents, and Excel files.
The biggest difference between tiers is therefore not simply what the Agent can do, but how much work it can handle.
For light users, ¥19 provides an inexpensive way to access AI productivity capabilities. For heavy users, the ¥128 tier provides substantially more capacity.
Qwen has also separated office productivity from video generation.
Office assistance follows a subscription model, while AI video generation uses a quota-based payment model. Users receive a daily allocation of free video credits, with additional credits available for separate purchase.
That creates a more flexible pricing structure: someone who occasionally generates a video does not necessarily need to purchase an expensive all-in-one membership.
The underlying strategy is straightforward:
Qwen is not primarily selling conversation. It is selling an Agent that can complete work.
You pay ¥19, and the Agent can help retrieve information, prepare presentations, organize documents, and handle routine workflows. Higher-priced plans simply expand the amount of work the Agent can perform.
It is essentially the traditional SaaS subscription model being rebuilt around AI Agents:
monthly payment + tiered capabilities + differentiated usage capacity.
Doubao’s Logic: Commission-Based Monetization Through Transactions
Doubao is taking a fundamentally different approach.
Beginning August 10, hotel transactions completed after users entered through Doubao and were redirected to Douyin Life Services were subject to a separate fee structure: 11.4% software service fee + 0.6% payment fee, for an effective rate of approximately 12%.
Previously, similar orders were treated as Douyin organic traffic and settled at a rate of roughly 8%.
The important distinction is that Doubao is not charging users simply for asking about hotels.
Instead, the platform monetizes the transaction that follows the recommendation.
That changes the commercial relationship:
Qwen charges for capability. Doubao charges for conversion.
For merchants, this is essentially a performance-based acquisition model. There is no need to pay upfront simply to obtain exposure; the cost occurs after an actual transaction is completed.
That makes the 12% fee easier to understand economically.
If AI recommendations generate genuinely incremental customers, the commission becomes a customer-acquisition cost that can potentially be justified by the additional revenue.
If, however, AI traffic simply redirects customers who would have purchased through Douyin anyway, the economics become much less attractive.
Doubao is therefore betting on a powerful combination:
AI traffic → recommendation → transaction → commission.
The more effectively the Agent closes this loop, the more valuable the traffic becomes.
The Fundamental Difference: Selling Capability vs. Selling Transactions
Put the two models side by side, and the distinction becomes obvious.
Qwen Sells “Capability”
Users pay for access to an intelligent productivity system.
Whether a particular task eventually generates commercial value is irrelevant to the subscription itself. The user has already paid for the capability.
That creates a relatively predictable revenue structure:
subscription → recurring revenue → predictable customer value.
Doubao Sells “Transactions”
Users can access the recommendation experience without directly paying a fee.
The merchant pays only after a transaction occurs.
This creates a different economic structure:
traffic → conversion → transaction → commission.
The upside can be much larger when transaction volume scales, but revenue is also more dependent on actual conversion.
In simple terms:
| Model | What is being monetized? | Who pays? | Revenue logic |
|---|---|---|---|
| Qwen | AI capability | User | Subscription |
| Doubao | Transaction conversion | Merchant | Commission |
| Token marketplace | Computing resources | Developer / Enterprise | Market-based pricing |
None of these models is inherently superior.
They simply fit different scenarios.
Productivity software naturally lends itself to subscriptions because users want continuous access to a reusable tool.
Transaction-oriented Agents naturally lend themselves to commissions because charging users for every search or recommendation would create friction, while charging merchants after a successful transaction is easier to justify.
But there is a third model emerging underneath both.
The Third Path: The Token Trading Market
Qwen represents subscription monetization.
Doubao represents transaction monetization.
But the infrastructure underneath AI applications is moving toward another possibility:
Token market pricing.
The concept behind Unexhub is to treat Tokens not merely as a usage meter, but as a market-based computing resource that can be priced, compared, routed, and traded.
Instead of forcing developers to accept a single vendor's fixed API price, a Token marketplace can aggregate multiple suppliers and allow prices to respond to supply and demand.
During periods of excess capacity, developers can potentially access lower prices.
During periods of high demand, suppliers can potentially capture higher returns.
For suppliers, the economic benefits are equally important.
Unexhub's model emphasizes real-time settlement, allowing suppliers to receive settlement based on actual usage rather than carrying long payment cycles.
For early suppliers, the platform also offers 0% commission during the initial participation period, lowering the cost of entering the marketplace and helping suppliers test new distribution channels without immediately giving up part of their revenue.
For developers and enterprises, the more direct benefit is price flexibility.
Rather than committing to one model provider and one fixed pricing schedule, users can compare and route workloads across multiple AI suppliers.
According to the platform's stated offering, developers can pay directly by Token and potentially access prices as low as 30% of prevailing market prices, depending on available supply and market conditions.
That turns a simple API purchase into something closer to a marketplace transaction.
Three Models, Three Layers of AI Economics
The differences can now be summarized in one framework:
1. Subscription: “Pay for Capability”
Qwen asks:
How much are you willing to pay every month for an AI Agent that works for you?
The core unit is the user subscription.
2. Commission: “Pay for Results”
Doubao asks:
How much value can the AI Agent generate from a successful transaction?
The core unit is the transaction.
3. Token Marketplace: “Let the Market Price Computing Power”
Unexhub asks a different question:
What is the market price of the computing resources required to deliver that capability right now?
The core unit becomes the Token.
These three models operate at different layers.
Qwen monetizes the application.
Doubao monetizes the transaction.
A Token marketplace attempts to monetize and efficiently allocate the underlying computing resource.
A New Competitive Landscape Is Taking Shape
The significance of Qwen’s ¥19 subscription and Doubao’s 12% commission goes beyond the numbers themselves.
They represent two different answers to the same question:
Where does the economic value of an AI Agent come from?
Qwen's answer is:
The Agent saves you time and completes work for you.
Doubao's answer is:
The Agent helps you complete a transaction.
The Token marketplace offers a third answer:
The Agent ultimately consumes computing resources, and those resources can themselves become market-priced.
This distinction could become increasingly important as AI Agents move from simple chat interfaces into autonomous workflows.
When an Agent can research information, create presentations, write code, operate software, compare products, contact customers, and complete transactions, the old idea of charging simply for "AI access" becomes increasingly difficult to sustain as the only pricing model.
What Does This Mean for Developers?
For developers, the most important change may be that model selection and pricing selection are becoming inseparable.
You may no longer ask only:
“Which model is the best?”
You may also need to ask:
Which model delivers the required quality?
Which provider offers the lowest current Token cost?
Is the workload latency-sensitive?
Is the provider reliable enough for production?
Can the workload automatically switch between models?
Can the same application access multiple suppliers without rewriting its integration layer?
This is where a Token aggregation and distribution infrastructure becomes increasingly valuable.
Unexhub positions itself as a computing-power trading center for global Token aggregation and distribution, bringing together AI suppliers of different scales, from major model providers such as Gemini, ChatGPT, Claude, DeepSeek, and GLM to smaller suppliers and specialized AI services.
The goal is not simply to provide another API gateway.
It is to create a marketplace where multiple suppliers can compete for enterprise and developer workloads through dynamic pricing, fast routing, and market-based distribution.
For developers, the potential value is straightforward: instead of paying a fixed price simply because a model comes from a particular vendor, they can pay according to actual Token usage and seek more competitive pricing when market supply allows it.
For suppliers, the proposition is equally direct: connect once, access a broader customer base, and receive settlement based on actual API usage.
The Bigger Question Is No Longer “How Much Does AI Cost?”
The emergence of subscription plans, transaction commissions, and Token marketplaces suggests that AI Agent economics is becoming more sophisticated.
¥19/month represents the price of continuous AI capability.
12% commission represents the price of successful transactions.
Token market pricing represents the price of the computing resources underneath those experiences.
These are not simply three competing price tags.
They are three different economic philosophies.
Subscription monetizes capability.
Commission monetizes results.
Token markets monetize computing resources through market pricing.
Qwen and Doubao have demonstrated that AI Agents can become commercial products in fundamentally different ways.
Unexhub is attempting to push the next layer further: making the computing resources consumed by those Agents more transparent, competitive, and market-driven.
The future of AI Agent economics may therefore not be about choosing one pricing model.
It may be about connecting all three.
中文版
19 元月费套餐 vs. 12% 抽佣:AI Agent 正在实验两种完全不同的赚钱方式
英中双语版|English Version 已在上方,中文内容见下文
2026 年 8 月,国内 AI Agent 的商业化,在同一天走上了两条完全不同的道路。
阿里千问推出办公助理付费会员,月费套餐从 19 元到 128 元,分为三个等级。
与此同时,字节跳动豆包开始对通过豆包入口跳转抖音生活服务并最终成交的酒店订单收取渠道服务费,综合费率约为 12%,其中包括 11.4% 的软件服务费和 0.6% 的支付手续费。
一个卖的是**“能力”**。
另一个赚的是**“交易”**。
两条路径最终都指向同一个问题:
AI Agent 到底应该怎么赚钱?
千问的逻辑:订阅制,卖“替你干活”的能力
千问办公助理采用三档会员体系:
高级版:19 元/月
精英版:49 元/月
旗舰版:128 元/月
对应的年付价格约为 200 元、568 元和 1499 元,额度则随着会员等级提升,分别达到约 2 倍、5 倍和 20 倍。
那么,19 元究竟能买到什么?
三档会员均可以使用 Qwen 旗舰模型,并获得本地与云端模式、手机远程操作、Skill 技能、连接器等能力,还能够直接生成网页、PPT、Word、Excel 等工作成果。
因此,会员之间真正拉开差距的,并不只是“能不能做”,而是:
一个月能够让 AI 替你完成多少工作。
对于轻度用户而言,19 元提供了一个非常低的尝试门槛;对于重度用户,128 元则提供更高的使用容量。
千问还有一个值得注意的设计:办公与视频生成采用两套计费体系。
办公助理采用订阅制,而 AI 视频生成则采用额度充值制。用户每天拥有一定的免费视频额度,超过之后再单独购买。
这种设计实际上更加灵活。
偶尔需要生成视频的用户,不必为了几个视频去购买一整套高价会员。
千问背后的商业逻辑因此非常清晰:
它卖的不是聊天次数,而是一个能够替用户完成工作的智能体。
19 元购买的是持续可用的 AI 办公能力;更高档位则对应更大的任务容量和更高的使用频率。
这本质上是传统 SaaS 订阅模式在 AI Agent 时代的一次升级:
按月付费 + 能力分层 + 使用额度分层。
豆包的逻辑:按交易抽佣,卖“交易撮合”的渠道
豆包选择的是完全不同的商业路径。
从 8 月 10 日开始,通过豆包入口进入抖音生活服务并最终完成酒店订单的交易,采用独立费率:
11.4% 软件服务费 + 0.6% 支付手续费 = 约 12% 综合费率。
此前类似订单按照抖音自然流量的规则结算,费率约为 8%。
关键区别在于:
豆包并不是向用户收取“找酒店”的费用。
它真正收费的是:
AI 推荐之后发生的交易。
这意味着商业关系发生了变化。
千问是在说:
“你付钱,我给你 AI 能力。”
豆包则是在说:
“你先使用 AI,只有产生交易之后,商家才为成交支付费用。”
对于商家而言,这更接近一种结果导向的获客模式。
商家无需为了获得 AI 推荐而提前支付费用,只有实际产生订单之后才承担渠道成本。
如果 AI 推荐真正带来了新增客户,那么 12% 的渠道服务费可以被视为一笔值得计算的获客成本。
但如果 AI 只是把原本属于抖音的流量重新分配了一遍,那么这 12% 的经济价值就需要重新评估。
所以豆包真正押注的是一条完整的商业链路:
AI 流量 → 智能推荐 → 用户决策 → 交易成交 → 平台抽佣。
只要这个闭环能够持续扩大,交易佣金的收入上限就会不断提高。
两种模式的本质差异:卖能力 vs. 卖交易
把千问与豆包放在一起,差异就非常明显了。
千问卖的是“能力”
用户付费购买的是一个可以持续使用的智能生产力工具。
无论某一次任务最终有没有创造商业价值,用户都已经为这项能力付费。
因此,它形成的是相对稳定的收入结构:
订阅 → 持续使用 → 经常性收入。
豆包卖的是“交易”
用户可以免费使用推荐能力,而商家只有在交易真正发生之后才支付费用。
因此形成的是另一种收入结构:
流量 → 转化 → 交易 → 抽佣。
简单来看:
| 模式 | 被变现的对象 | 谁付钱 | 收入逻辑 |
|---|---|---|---|
| 千问 | AI 能力 | 用户 | 订阅 |
| 豆包 | 交易转化 | 商家 | 抽佣 |
| Token 交易市场 | 算力资源 | 开发者 / 企业 | 市场定价 |
三者并没有谁绝对更好。
它们只是对应不同的商业场景。
办公生产力天然适合订阅,因为用户需要长期使用一个稳定的工具。
而交易型 Agent 天然适合佣金,因为让用户为每一次搜索或推荐付费并不现实,让商家在成交后支付费用则更加合理。
但在这两种模式之外,还有第三条路径正在形成。
第三条路径:Token 交易市场
千问代表的是订阅商业化。
豆包代表的是交易佣金商业化。
而 AI 应用背后的算力基础设施,则正在出现另一种可能:
Token 市场定价。
Unexhub 的核心思路,是不再把 Token 仅仅看作 API 的计量单位,而是进一步将其变成可以定价、比价、竞价、路由和交易的算力资源。
传统模式下,开发者往往只能接受某一家模型厂商公布的固定价格。
而 Token 聚合交易市场试图改变这一点:
将不同模型、不同供应商、不同算力资源集中到同一个市场中,让价格根据供需关系动态变化。
在算力供应充足的时候,开发者有机会获得更低价格。
在需求高峰时,供应商则有机会获得更高收益。
对于供应商而言,即时结算同样具有实际价值。
Unexhub 强调按照实际调用进行结算,让供应商能够减少传统渠道中长期账期、提前垫资等资金压力。
与此同时,早期入驻供应商可以享受最低 0 佣金的参与权益,从而降低进入市场、测试新渠道和获取客户的成本。
对于开发者和企业而言,更直接的价值则是:
拥有更多价格选择。
不再必须绑定某一家模型厂商,也不必长期接受单一价格体系,而是可以根据模型能力、实时价格和业务需求选择更加合适的供应商。
按照平台目前的服务定位,开发者可以按 Token 实际用量付费,在市场供给允许的情况下,最低可享受市场价格 3 折左右的调用成本。
这让原本单纯的 API 采购,开始具备更多市场交易的特征。
三种模式,三层 AI 经济学
如果把三种模式放在一起,可以得到一个更加清晰的框架。
1. 订阅制:“为能力付费”
千问回答的是:
你愿意每个月花多少钱,让一个 AI Agent 持续替你工作?
核心计量单位是:
用户订阅。
2. 佣金制:“为结果付费”
豆包回答的是:
AI Agent 能够帮助平台和商家创造多少实际交易价值?
核心计量单位是:
交易。
3. Token 交易市场:“让市场决定算力价格”
Unexhub 关注的则是另外一个问题:
完成这些 AI 任务所需要的算力资源,在当前市场上到底应该值多少钱?
核心计量单位变成:
Token。
三种模式其实处在不同层级。
千问变现的是应用能力。
豆包变现的是交易结果。
Token 交易市场则试图进一步优化和配置底层算力资源。
一个新的竞争格局正在形成
千问的 19 元会员与豆包的 12% 抽佣,真正值得关注的并不是两个数字本身。
它们代表的是对同一个问题的两种不同答案:
AI Agent 的经济价值究竟来自哪里?
千问的答案是:
AI Agent 帮你节省时间、完成工作。
豆包的答案是:
AI Agent 帮你完成交易、创造商业转化。
而 Token 交易市场提供了第三种答案:
AI Agent 最终消耗的是算力,而算力本身也可以成为市场定价的对象。
随着 AI Agent 从聊天工具逐渐走向复杂工作流,这一点会越来越重要。
当一个 Agent 可以检索信息、制作 PPT、写代码、操作软件、分析数据、寻找客户,甚至直接完成交易时,仅仅按照“AI 使用权”收费,可能已经不再是唯一可行的商业模式。
对开发者而言,真正的变化是什么?
未来,开发者考虑的可能不再只是:
“哪个模型最好?”
而是需要同时考虑:
哪个模型能够满足业务质量要求?
哪个供应商当前 Token 价格更低?
任务是否对延迟敏感?
供应商是否足够稳定?
能否根据业务自动切换模型?
如果供应商涨价,是否可以快速切换?
一个应用能否同时接入多个供应商,而无需重写整个系统?
这正是 Token 聚合与分发基础设施开始体现价值的地方。
Unexhub 将自身定位为全球 Token 聚合与分发的算力交易中心,汇聚不同规模的 AI 供应商,包括 Gemini、ChatGPT、Claude、DeepSeek、GLM 等大型模型供应商,同时连接个人和中小企业提供的专业 AI 模型与算力资源。
它的目标并不只是再提供一个 API 网关。
更重要的是,希望构建一个能够让不同供应商进行市场竞争的分发体系,通过:
动态定价 + 快速路由 + 多供应商竞争 + Token 按量结算
让企业和开发者拥有更多选择。
对于开发者而言,这意味着不必仅仅因为“某个模型属于某个厂商”,就被迫接受固定价格,而可以根据实时供应、模型能力和市场价格选择更合适的调用方案。
对于供应商而言,则可以通过一次接入进入更广泛的开发者与企业市场,并按照实际 API 调用获得对应结算。
真正的问题已经不再是“AI 到底多少钱”
订阅、佣金与 Token 交易市场的同时出现,说明 AI Agent 的商业经济学正在变得更加复杂。
19 元/月,代表的是持续 AI 能力的价格。
12% 抽佣,代表的是成功交易的价格。
Token 市场价格,代表的则是支撑这些 AI 能力运行的底层算力成本。
它们并不是三个简单的价格标签。
而是三种不同的商业哲学:
订阅制,变现能力。
佣金制,变现结果。
Token 市场,利用市场机制重新定价算力。
千问和豆包已经证明,AI Agent 可以通过完全不同的方式实现商业化。
而 Unexhub 试图继续向底层推进一步:
让 AI Agent 背后的算力资源变得更加透明、更具竞争性,也更加市场化。
未来的 AI Agent 经济,可能并不是三选一。
真正的趋势,也许是让这三种模式最终连接起来:
应用通过订阅获得收入,交易通过佣金实现变现,而底层算力则通过 Token 市场进行更高效的定价与分配。
这可能才是 AI Agent 商业化真正走向成熟之后的完整经济闭环。