AI agent app store 大洗牌:千问/豆包免费版与付费版差在哪?19 元到 128 元档位全对比

Two leading Chinese AI applications have now moved beyond the free-only model. From 19 yuan to 500 yuan, the price gap is more than 26×.


From 19 Yuan to 500 Yuan: The AI Agent App Market Is Running a Pricing Experiment

Bilingual Edition · English First / 中文版见下方

In August 2026, China’s AI application market is entering a new phase: AI Agent products are beginning to move beyond the free model and experiment with real commercial pricing.

On the evening of August 10, Alibaba’s Qwen App quietly introduced paid membership plans. Its Office Assistant is divided into three tiers: 19 yuan/month for Advanced, 49 yuan/month for Elite, and 128 yuan/month for Ultimate. AI video generation is billed separately through credits, starting at 26 yuan for 10 credits.

Just six weeks earlier, on June 24, ByteDance’s Doubao had already crossed the subscription threshold. Its plans range from 68 yuan/month for Standard to 200 yuan/month for Enhanced and 500 yuan/month for Advanced.

Two leading Chinese AI applications have now moved beyond the free-only model.

From 19 yuan to 500 yuan, the price gap is more than 26×.

But this is not simply a price war. It is a large-scale industry experiment around a fundamental question:

How should AI Agents actually be priced?


Qwen's Strategy: Low-Cost Entry, Separate Pricing for Office and Video

Qwen’s pricing strategy has a clear market-penetration flavor.

At 19 yuan per month, the entry price is low enough to make the decision almost frictionless. The Elite plan costs 49 yuan, while the Ultimate plan costs 128 yuan.

All three membership tiers provide access to Qwen’s latest flagship model capabilities, together with features such as local and cloud modes, remote smartphone control, Skills, and Connectors.

The primary difference is usage capacity, with quotas scaling by roughly 2×, 5×, and 20× across the tiers.

The more interesting part is Qwen’s decision to separate office assistance and AI video generation into two billing systems.

Office assistance follows a subscription model, while AI video generation uses a credit-based system. Users receive a certain amount of free video-generation credits each day, with additional usage available through paid credit packages.

This creates a more flexible “subscription + pay-as-you-go” model.

A light video user does not necessarily need to purchase an expensive monthly package simply because they occasionally want to generate a video.

The underlying strategy is straightforward:

Lower the entry barrier, then let users pay according to what they actually consume.


Doubao's Strategy: Sell Productivity, Not Just Quota

Doubao is taking a different route.

Its pricing starts at 68 yuan and goes all the way to 500 yuan per month. The Standard, Enhanced, and Advanced plans progressively increase usage quotas, while verified university students can receive a special Standard-plan price of 38 yuan/month.

But the real difference between Doubao’s free and paid versions is not simply the number of credits.

It is the underlying model and Agent capability.

Free users rely on a lower-tier model, while paid users gain access to a more capable Pro model.

The distinction can be summarized simply:

The free version helps you chat. The paid version is designed to help you get work done.

The Agent can break down complex objectives, interact with local files, search the web, use specialized tools, and automate multi-step workflows.

In other words, Doubao is not primarily selling credits.

It is selling the value proposition of:

“Let AI take over part of the work for you.”

For a user paying 68 yuan a month, the value is not merely more conversations. It is the possibility of having an Agent independently handle tasks such as PPT formatting, data-report generation, information research, and other multi-step office workflows.


The Real Difference Between Free and Paid AI

Put the two products side by side, and the difference between free and paid AI can be summarized across three dimensions.

1. Model Capability

Paid Qwen members can access flagship-model capabilities with higher usage limits.

Doubao takes a more explicit approach by differentiating the underlying model available to free and paid users.

The message is clear:

Premium pricing increasingly buys access to a higher capability ceiling.

2. Agent Capability

This may be the most important distinction.

Qwen’s Office Assistant can connect with tools such as calendars, email, notes, computers, browsers, Skills, and other services to execute multi-step tasks.

Doubao’s productivity Agents similarly focus on decomposing complex objectives, interacting with local files, and scheduling batch tasks.

The difference is no longer simply:

“How many questions can I ask?”

It is becoming:

“How much work can the AI actually complete for me?”

3. Quota and Concurrency

Free plans can provide access to AI capabilities, but paid plans significantly expand usage capacity and concurrency.

For users who occasionally experiment with AI, free access may be enough.

For users who rely on Agents throughout the workday, however, quota itself becomes part of the product value.


From 19 Yuan to 500 Yuan: Who Is Each Product Really Pricing For?

Qwen’s 19-yuan entry plan is roughly 72% cheaper than Doubao’s 68-yuan entry plan.

That does not necessarily mean Qwen’s underlying inference costs are several times lower.

The difference is primarily a difference in market positioning.

Qwen is competing for the user’s first paid conversion.

At 19 yuan, the psychological barrier is extremely low. Users can subscribe almost without deliberation, experience the product, and potentially upgrade later.

Doubao, meanwhile, is anchoring itself closer to the traditional productivity-software price range.

A 68-yuan starting point communicates a different message:

This is not just an AI chatbot. It is a productivity tool.

Neither strategy is inherently right or wrong.

They simply reflect different assumptions about where the AI Agent market currently stands.

Qwen appears to be betting that the market is still in the user-education and penetration phase.

Doubao is betting that users are already willing to pay for measurable productivity gains.


But Token Markets Introduce a Completely Different Pricing Logic

Qwen and Doubao are conducting pricing experiments at the application-store level.

The manufacturer sets the price.

The user decides whether to subscribe.

A Token trading market works differently.

The core idea behind Unexhub is to transform Tokens from a simple API billing unit into a market-priced computational resource that can be compared, routed, and traded.

Instead of forcing users to accept a fixed vendor price, an aggregation and distribution marketplace can connect multiple model suppliers and allow API prices to respond to supply and demand.

This creates several potential benefits.

Dynamic Pricing

During periods of excess capacity, developers may be able to access models at lower prices.

During periods of high demand, suppliers can potentially earn higher returns.

The price therefore becomes a reflection of real-time market conditions, rather than simply a fixed number on a vendor’s pricing page.

Real-Time Settlement

For smaller AI suppliers, cash flow can be just as important as customer acquisition.

Unexhub’s real-time settlement model is designed around the principle of:

Supply capacity → generate usage → settle revenue.

Suppliers do not need to rely solely on long payment cycles, while developers gain access to a broader supply pool.

Early-Stage Zero Commission

For early suppliers joining the platform, Unexhub offers commission rates starting from 0%.

This is more than a promotional discount.

It lowers the cost of entering the marketplace, allowing smaller suppliers to test demand and build distribution channels without immediately sacrificing revenue to platform commissions.

For developers, the value is on the other side of the marketplace: more suppliers competing for demand can create more room for price discovery and lower-cost access.


The Real Change: From Subscription Pricing to Market Pricing

If this model develops successfully, a 19-yuan Qwen membership or a 500-yuan Doubao membership may no longer represent the entire pricing story.

They would become reference points within a much larger AI service economy.

Developers could potentially choose between:

  • different models;

  • different suppliers;

  • different API routes;

  • different usage periods;

  • different price levels.

The fundamental question would change from:

“Which AI subscription should I buy?”

to:

“Which model can deliver the best performance-to-cost ratio right now?”

That is a much more dynamic pricing mechanism.


A Market That Is Still Taking Shape

The move from free AI to paid Agents is only the beginning.

As more AI applications introduce subscriptions, usage quotas, credit systems, and premium Agent capabilities, the industry is moving from “Who has the best chatbot?” toward a much broader competition around:

  • Agent productivity;

  • multimodal generation;

  • usage quotas;

  • model capability;

  • inference cost;

  • routing efficiency;

  • and price-performance ratios.

The 19-yuan Qwen membership and 500-yuan Doubao plan are therefore more than two price tags.

They are signals that AI Agents are becoming commercial productivity infrastructure.

But the emergence of Token aggregation and trading platforms introduces another possibility.

Instead of every AI vendor deciding the final price independently, multiple suppliers can compete for the same developer and enterprise demand.

That is where Unexhub is positioning itself.

Unexhub aims to build a Token aggregation and distribution computing-power trading center, bringing together AI suppliers of different sizes—from major model providers such as Gemini, ChatGPT, Claude, DeepSeek, and GLM to high-quality models offered by smaller teams and independent suppliers.

Through multi-supplier aggregation, competitive pricing, and millisecond-level routing, developers can access different AI models through a unified channel rather than opening separate accounts and managing separate API balances.

More importantly, the platform’s pay-per-token model means developers pay for actual usage rather than being forced into a single fixed subscription.

For eligible routes, Unexhub says developers can access AI capacity at as low as 30% of prevailing market prices, depending on supplier, model, route, and real-time market conditions.

For suppliers, the other side of the marketplace is equally important: one-click channel integration and real-time settlement of usage-based revenue, making it easier for AI capacity providers to turn available inference resources into actual distribution and revenue.

The AI Agent market may have started with subscriptions. The next stage may be defined by Token-level market pricing.

From 19 yuan to 500 yuan, today’s price tags show how aggressively AI companies are testing user willingness to pay.

But tomorrow’s competition may be less about which company sets the lowest subscription price and more about which marketplace can connect the right model, supplier, route, and price at the right moment.

The AI Agent app store is being rewritten.

And the next chapter may not be written by any single model provider—but by the economics of the Token market itself.


中文版

从 19 元到 500 元:AI Agent 应用商店正在经历一场定价实验

双语版说明:英文内容已完整呈现在上方,以下为中文版。

2026 年 8 月,国内 AI 应用市场正在进入一个新的阶段:

AI Agent 开始真正走出“免费体验”,进入商业化定价时代。

8 月 10 日晚,阿里千问 App 悄然推出付费会员服务。办公助理分为三档:高级版 19 元/月、精英版 49 元/月、旗舰版 128 元/月。AI 视频生成则采用独立额度体系,最低 26 元可购买 10 个额度。

六周前的 6 月 24 日,字节跳动豆包率先跨过收费线:标准版 68 元/月、加强版 200 元/月、高级版 500 元/月。

两款国内头部 AI 应用,先后跨过了免费门槛。

从 19 元到 500 元,价格相差超过 26 倍。

但这并不只是一场简单的价格战。

它实际上正在回答一个更重要的问题:

AI Agent 到底应该如何定价?


千问的算盘:低价渗透,办公与视频分开收费

千问的定价策略带着明显的市场渗透色彩。

19 元/月的入门价格非常低,几乎不会形成明显的购买决策门槛。精英版 49 元/月,旗舰版 128 元/月。

三档会员均可以使用千问旗舰模型能力,并支持本地与云端模式、手机远程操作、Skill 技能以及连接器等能力。

不同版本的核心区别之一,在于使用额度,随着会员等级提升,额度大致扩大至 2 倍、5 倍和 20 倍。

更值得关注的是,千问并没有把所有 AI 能力塞进一个固定月费里,而是把办公助理和 AI 视频生成拆成两套计费体系。

办公助理采用订阅制,视频生成则采用额度制。

这种“订阅 + 按量付费”的模式其实更加灵活。

对于偶尔生成视频的轻度用户来说,不需要为了偶尔使用一次视频能力,就购买更高价格的完整会员。

千问的商业逻辑非常清晰:

先用低门槛吸引用户,再让不同需求的用户为实际需要的能力付费。


豆包的算盘:卖生产力,而不只是卖额度

豆包选择的是另一条路线。

68 元起步,500 元封顶。

标准版、加强版、高级版逐级提高使用额度,同时,经过大学生认证后,标准版还可以享受 38 元/月的特惠价格。

但豆包付费版真正的差异,并不只是“多给多少额度”。

更重要的是:

底层模型和 Agent 能力发生了变化。

免费用户使用相对基础的模型,而付费用户可以获得更高能力的 Pro 模型。

因此,两者之间的区别可以简单概括为:

免费版帮助你聊天,付费版开始帮助你干活。

付费 Agent 可以自主拆解复杂目标、处理本地文件、搜索互联网、调用专业工具,并完成多步骤工作流。

所以豆包真正销售的并不是单纯的 Token 或额度。

它销售的是:

“让 AI 替你完成一部分工作”的生产力价值。

用户支付 68 元购买的,不只是更多对话次数,而是让 Agent 帮助完成 PPT 排版、数据报告、信息检索以及其他复杂办公流程的可能性。


免费版与付费版,真正差在哪里?

把两款产品放在一起,免费与付费 AI 的差异,大致可以归结为三个维度。

1. 模型能力

千问付费会员可以获得旗舰模型能力,同时拥有更高的使用额度。

豆包则进一步通过不同底层模型,对免费和付费用户进行能力区隔。

这意味着:

未来的付费会员,购买的不只是更多次数,更可能是更高的能力上限。

2. Agent 能力

这可能才是最核心的变化。

千问办公助理可以连接备忘录、日历、邮箱、电脑、浏览器以及各种 Skill 和服务,完成多步骤任务。

豆包的生产力 Agent 同样强调复杂目标拆解、本地文件操作以及定时批量任务。

因此,AI 产品的竞争问题正在发生变化。

过去用户问的是:

“我还能和 AI 聊多少次?”

现在用户越来越关心的是:

“AI 到底能替我完成多少工作?”

3. 额度与并发

免费版依然能够使用 AI 能力,但付费版会明显扩大额度和并发能力。

对于偶尔尝试 AI 的用户来说,免费版可能已经足够。

但对于每天依赖 Agent 工作的用户来说:

额度本身已经成为产品价值的一部分。


从 19 元到 500 元:到底在为谁定价?

千问 19 元的入门价格,比豆包 68 元的起步价低约 72%。

这并不意味着千问的实际推理成本一定比豆包低数倍。

更大的区别,其实来自市场定位。

千问争夺的是用户的第一次付费转化。

19 元的心理门槛足够低,用户几乎不需要太多考虑就可以购买、体验,再逐渐向更高版本升级。

豆包则更接近传统生产力软件的定价区间。

68 元的起步价实际上是在向用户传递另一个信号:

这不是一个普通聊天机器人,而是一款生产力工具。

两种策略没有绝对的对错。

它们只是对 AI Agent 市场所处阶段的不同判断。

千问更像是在押注:

市场仍处于用户教育和快速渗透阶段。

而豆包则更像是在押注:

用户已经开始愿意为明确的生产力提升付费。


但 Token 交易市场带来了另一套定价逻辑

千问和豆包正在进行的是应用商店层面的定价实验。

厂商制定价格。

用户决定是否订阅。

而 Token 交易市场的逻辑完全不同。

Unexhub 的核心思路,是把 Token 从简单的 API 计量单位,进一步变成一种可以比价、路由和交易的算力资源。

开发者不再只能接受某一家厂商固定的 API 价格,而可以通过聚合分发平台连接多个模型和供应商,让价格随着市场供需变化。

这会带来几个重要变化。

动态定价

在算力供给充足的时段,开发者有机会获得更低的价格。

在需求高峰期,供应商则有机会获得更高的收益。

价格因此不再只是厂商官网上的一个固定数字,而开始反映实时市场供需。

即时结算

对于中小 AI 供应商来说,资金周转与获客同样重要。

Unexhub 的即时结算逻辑强调:

供应多少 → 产生多少调用 → 获得对应结算。

供应商不需要完全依赖漫长的账期,开发者则可以获得更丰富的算力供应。

早期零佣金

对于早期入驻的供应商,Unexhub 提供最低 0 佣金的入驻激励。

它的意义并不只是简单的优惠。

对于供应商而言,较低甚至 0 佣金可以降低进入市场的成本,让中小供应商能够更轻量地测试市场需求、拓展销售渠道。

对于开发者而言,更多供应商进入同一个市场,也意味着更充分的竞争、更丰富的模型选择,以及更大的价格发现空间。


真正的变化:从订阅定价走向市场定价

如果 Token 市场进一步成熟,那么千问 19 元/月、豆包 500 元/月这样的会员价格,可能不再是 AI 服务定价的全部。

它们更可能成为一个更大 AI 服务市场中的价格参考点。

开发者未来可以比较:

  • 不同模型;

  • 不同供应商;

  • 不同 API 路由;

  • 不同使用时段;

  • 不同价格水平。

于是问题也会发生变化。

过去用户问:

“我应该买哪家的 AI 会员?”

未来可能变成:

“现在什么模型的性能价格比最高?”

这才是真正动态化的定价逻辑。


一个正在成形的新市场

从免费 AI 到付费 Agent,只是第一步。

随着越来越多 AI 应用推出订阅、额度、积分以及高级 Agent 能力,行业竞争也正在从:

“谁的聊天能力更强?”

逐渐转向:

  • Agent 生产力;

  • 多模态生成;

  • 使用额度;

  • 模型能力;

  • 推理成本;

  • 路由效率;

  • 性价比。

因此,19 元和 500 元并不只是两个价格。

它们代表着一个重要信号:

AI Agent 正在从消费级应用逐渐成为商业生产力基础设施。

而 Token 聚合与交易平台的出现,则提供了另一种可能。

未来不一定是每一家 AI 厂商分别决定最终价格,而可能是多个供应商围绕同一批开发者和企业需求进行竞争。

这正是 Unexhub 希望切入的方向。

Unexhub 定位于Token 聚合与分发的算力交易中心,汇聚不同规模的 AI 供应商,既包括 Gemini、ChatGPT、Claude、DeepSeek、GLM 等主流模型,也包括个人开发者、中小企业提供的优质 AI 模型。

通过多供应商聚合、市场竞价以及毫秒级路由,开发者无需分别注册多个平台、维护多个 API 账户和余额,即可通过统一渠道访问不同模型能力。

更重要的是,Unexhub 采用按 Token 实际使用量付费的模式,让开发者不必被固定套餐完全绑定。

根据供应商、模型、路由以及实时市场情况不同,Unexhub 宣称部分线路的开发者使用成本最低可达到市场价格的 3 折。

对于供应商而言,平台的另一侧同样重要:

一键接入渠道 + 调用量实时结算分成。

这意味着 AI 算力供应商可以更便捷地把闲置推理资源转化为实际调用和收入,同时通过市场竞争获得更多分发机会。

💡 AI Agent 的商业模式可能从订阅开始,但 Token 市场可能决定它的下一阶段。

从 19 元到 500 元,今天的价格标签,是 AI 厂商对用户付费意愿的一次次测试。

但下一阶段的竞争,也许不再只是:

“哪家厂商的会员更便宜?”

而会变成:

“谁能够在正确的时间,用正确的模型、正确的供应商和正确的价格,为用户匹配最合适的算力?”

AI Agent 应用商店正在被重新定义。

而下一章,可能不再由某一家模型厂商单独书写,而是由 Token 市场本身的经济规律来书写。