字节20亿元对腾讯5亿美元:AI办公入口之争ByteDance's 2 Billion Yuan vs Tencent's 500 Million Dollars: The Battle for the AI Office Gateway


AI free token/AI免费积分

字节20亿元对腾讯5亿美元:AI办公入口之争

本文来源:https://www.jiemian.com/article/15357591.html
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

豆包工作上线三个月,DAU突破800万,企业ARR超过20亿元。同一天,WorkBuddy披露DAU超过1000万,日均Token消耗100万亿,近七天收入年化折算超过5亿美元。

放在一年前,这组数字只有一个读法:速度竞赛。谁跑得快,谁先撞线。但到了2026年下半年,数字本身退到其次。真正值得拆的,是两组数字背后两套完全不同的入口逻辑——以及它们对“AI办公究竟在争什么”给出的两个答案。

孤立地看豆包工作的ARR,会失真。它得放进飞书的坐标系里读。

飞书2025年ARR已超30亿元,2026年二季度ARR同比增长超100%。豆包工作三个月做到20亿元ARR,增速惊人,但其中相当一部分来自飞书既有客户的升级与迁移,而非全新的企业采购。

字节的路径很清楚:不另起炉灶造一个AI原生的企业入口,而是把豆包嵌进飞书已经沉淀好的文档、知识库、会议与协作场景。飞书过去几年在中大型客户那里攒下的组织关系、工作流数据和付费习惯,就是豆包工作最短的商业化路径。

这条路的好处是ARR的“厚度”。企业年度经常性收入天生比消费订阅更稳、更可预测;一旦嵌进工作流,替换成本就是事实上的护城河。

代价同样清楚:豆包工作的增长节奏,很大程度上被飞书的客户结构定义。飞书擅长的是知识密集、协作密集的组织,而这在中国企业版图里并不是最大公约数。20亿元ARR之后更该追问的是——在飞书还没覆盖的行业和规模段里,豆包工作能不能独立跑出增量?

它的前身CodeBuddy是面向开发者的内部效率工具,早期团队直接驻场企业客户,在会议和部署里打磨产品。等产品降低门槛、转向非技术用户的办公场景,接上腾讯文档、QQ邮箱、腾讯新闻这些消费级触点,才真正进入规模化轨道。

腾讯披露的1000万DAU和100万亿日均Token,本质上是消费端频次的胜利:用户先在个人工作场景里形成高频依赖,企业采购意愿随后被“倒逼”出来。

业内一个判断是:“WorkBuddy是用户先用起来,再倒逼组织关注;豆包和千问则是通过已有B端产品直接切入组织。”这解释了为什么WorkBuddy在QuestMobile 9月办公Agent月活榜上以3235.3万居首、季度环比增长647.1%,而豆包工作在同口径下的月活,远不及它DAU暗示的体量。

WorkBuddy的流量逻辑更像消费产品。它的壁垒不在企业合同里,在用户日复一日的操作习惯里。

但“倒逼”的脆弱也在这里:消费端高频,不会自动变成组织级采购。腾讯说中国百强企业已有60%成为WorkBuddy客户,这个数字的含金量取决于“客户”怎么定义——是完成了企业级部署与付费,还是仅仅有一批活跃的个人用户。从个人订阅到企业合同,中间那道转化漏斗,是WorkBuddy下一阶段必须交出的答卷。

三家在做同一件事:把第三方模型纳入自己的产品体系。

WorkBuddy整合了DeepSeek、智谱GLM、Kimi和MiniMax;豆包工作在自研豆包模型之外接入Kimi、GLM和DeepSeek;千问办公在11月中旬一次性上线10款模型,除Qwen系列外同样覆盖主流第三方。

放在一年前,这会被读作“不够专注”。到了2026年,它只说明一个事实:在办公场景里,底层模型的能力差异,已经不是用户选产品的主要变量。

模型不再是差异化来源,竞争就回落到三个更“重”的维度:渠道、客户关系、产品系统。

字节的牌是飞书已经建好的企业协作系统,AI只是嵌进这套系统的一层能力。阿里的牌是钉钉的销售体系和企业流程数据,千问办公更接近“在组织内部执行任务”的Agent。腾讯独有的牌,是微信生态连着的消费者关系与使用频次——WorkBuddy从C端切入,本质是把腾讯最擅长的“高频入口”能力,往工作场景里延伸。

但三家最近的收拢动作,也各自暴露了焦虑。字节把飞书产品团队并入豆包体系,是和“协同办公唱主角”的时代告别;阿里把QoderWork、悟空、MuleRun三条Agent赛道捏成千问办公,是追赶中的集中火力;腾讯把QClaw划入WorkBuddy管理体系,动作最小,方向同样明确。

共识只有一个:分散的实验没意义了。入口,只能有一个。

20亿元ARR和5亿美元年化收入,表面看都是“AI办公能赚钱”的证据。拆开看,它们回答的是两个问题。

豆包工作的ARR回答的是:企业愿不愿为AI工作流付费。三个月做到20亿元,说明在飞书已覆盖的组织里,AI能力是被当作协作基础设施的延伸采购的,而不是一个要单独论证ROI的新品类。艾媒咨询白皮书显示,企业付费中约一半采用“基础订阅+超额用量”的混合模式——这意味着收入上限取决于AI在组织内部被用得多深,而不是卖了几个席位。

WorkBuddy的DAU和Token消耗回答的是另一个问题:个人工作流能不能产生足够高的频次,高到AI成为日常操作里剥不掉的一层。100万亿日均Token给出了肯定的答案。

但“高频”本身不是商业模式。从个人高频到企业级收入,中间隔着订阅转化、组织采购、合规部署、IT集成四道关。腾讯用“最近七天年化”来呈现收入规模,传播上有效,财务审慎性上却留了余地——月度是行业公认的最短合理年化周期,七天数据的波动性不言自明。

两个问题并不互斥,但答案指向的组织能力不同。答好第一个,靠企业销售体系、客户成功和行业解决方案;答好第二个,靠产品体验、增长机制和生态触点。两个都答好,比在任意单一路径上跑出漂亮数据难得多。

WorkBuddy在腾讯财报里的位置值得留意。

2026年三季度,腾讯营收2073.36亿元,同比增长8%,非IFRS净利润同比下降6%;但云服务收入同比增长35%,腾讯明确把WorkBuddy的需求列为驱动力之一。在整体增速放缓的背景下,AI办公产品第一次进入核心财务叙事。这本身就是信号:WorkBuddy不再是一个“尝试性产品”,它扛着腾讯在AI时代从消费互联网向企业服务延伸的期待。

马化腾11月内部讲话里那句“陆战已经变成海战,陆军必须变成海军”,放进这个语境并不夸张。腾讯在AI原生应用第一梯队里的缺位是结构性的:QuestMobile 9月AI原生App月活榜前三分别是豆包4.19亿、千问1.58亿、DeepSeek 1.23亿,腾讯无一入围。WorkBuddy的3235万办公Agent月活,是腾讯在“工作场景”里为数不多能对位的资产。

它必须从“跑得最快的办公Agent”,变成“收得住收入的企业平台”,才能补上腾讯AI版图最薄的一块。

对整个赛道来说,同一天的两组数据给了个清晰的观察框架:豆包工作检验企业付费的可持续性,WorkBuddy检验消费端频次向组织采购的转化效率。两个检验,都还没到终局。

AI办公最终争的,不是一个新品类,而是个人工作流与企业工作流的交汇处。在这儿,DAU和ARR只是某一刻的快照。真正决定格局的,是谁能让AI同时读懂一个人的操作习惯和一个组织的运行规则,再把两者接进同一条工作流。

发布数据的日子会过去。入口之争,才刚进入要证明“留量”而非“流量”的阶段。

ByteDance's 2 Billion Yuan vs Tencent's 500 Million Dollars: The Battle for the AI Office Gateway

本文来源:https://www.jiemian.com/article/15357591.html
若有版权问题,请来信告知:13951431913#139.com 或在右方提交评论。

Three months after its launch, Doubao Work surpassed 8 million DAU and its enterprise ARR exceeded 2 billion yuan. On the same day, WorkBuddy(了解更多) disclosed over 10 million DAU, daily average token consumption of 100 trillion, and annualized revenue over the past seven days exceeding US$500 million.

A year ago, these numbers would have had only one reading: a speed race. Whoever runs faster crosses the line first. But in the second half of 2026, the numbers themselves take a back seat. What's truly worth dissecting are the two completely different entry-point logics behind the two sets of numbers—and the two answers they give to the question of “what exactly is the AI office market competing for.”

Looking at Doubao Work's ARR in isolation is misleading. It has to be read within the coordinate system of Feishu (Lark).

Feishu's ARR already exceeded 3 billion yuan in 2025, and its ARR grew over 100% year-on-year in Q2 2026. Doubao Work reaching 2 billion yuan ARR in three months is astonishing growth, but a considerable portion of it comes from upgrades and migration of Feishu's existing customers, rather than brand-new enterprise purchases.

ByteDance's path is clear: rather than starting from scratch to build an AI-native enterprise entry point, it embeds Doubao into the documents, knowledge bases, meetings, and collaboration scenarios that Feishu has already cultivated. The organizational relationships, workflow data, and payment habits Feishu has accumulated among mid-to-large customers over the past few years are the shortest commercialization path for Doubao Work.

The advantage of this path is the “thickness” of the ARR. Enterprise annual recurring revenue is inherently more stable and predictable than consumer subscriptions; once embedded into a workflow, switching cost becomes a de facto moat.

The cost is equally clear: Doubao Work's growth rhythm is largely defined by Feishu's customer structure. Feishu excels with knowledge-intensive, collaboration-intensive organizations, which are not the greatest common denominator in China's corporate landscape. After 2 billion yuan ARR, the more important question to ask is—in the industries and size segments Feishu hasn't yet covered, can Doubao Work independently generate incremental growth?

Its predecessor, CodeBuddy, was an internal productivity tool for developers; in the early days the team was stationed directly at enterprise customer sites, polishing the product through meetings and deployments. Only after the product lowered its barrier to entry, shifted to office scenarios for non-technical users, and connected with consumer-grade touchpoints like Tencent Docs, QQ Mail, and Tencent News did it truly enter a track of scale.

Tencent's disclosed 10 million DAU and 100 trillion daily average tokens are essentially a victory of consumer-side frequency: users first form a high-frequency dependency in personal work scenarios, and enterprise purchasing intent is then “forced” out.

An industry assessment goes: “With WorkBuddy, users adopt it first, then force organizations to take notice; Doubao and Qwen cut directly into organizations through existing B2B products.” This explains why WorkBuddy topped QuestMobile's September office Agent monthly-active ranking with 32.353 million, up 647.1% quarter-on-quarter, while Doubao Work's monthly actives under the same metric are far below the scale its DAU implies.

WorkBuddy's traffic logic is more like that of a consumer product. Its moat lies not in enterprise contracts, but in users' day-to-day operating habits.

But the fragility of “forcing” is also here: high consumer-side frequency doesn't automatically become organizational-level procurement. Tencent says 60% of China's top 100 enterprises have become WorkBuddy customers; the value of this number depends on how “customer” is defined—whether it means completed enterprise-grade deployment and payment, or merely a group of active individual users. The conversion funnel between individual subscriptions and enterprise contracts is the answer WorkBuddy must deliver in its next phase.

All three are doing the same thing: incorporating third-party models into their own product systems.

WorkBuddy has integrated DeepSeek, Zhipu GLM, Kimi, and MiniMax; Doubao Work, besides its self-developed Doubao model, has connected Kimi, GLM, and DeepSeek; Qwen Office launched 10 models at once in mid-November, covering mainstream third parties in addition to the Qwen series.

A year ago, this would have been read as “not focused enough.” By 2026, it only illustrates one fact: in office scenarios, differences in the capabilities of underlying models are no longer the main variable in users' product choices.

With models no longer a source of differentiation, competition falls back to three “heavier” dimensions: channels, customer relationships, and product systems.

ByteDance's card is the enterprise collaboration system Feishu has already built, with AI merely a layer of capability embedded in it. Alibaba's card is DingTalk's sales system and enterprise process data, with Qwen Office closer to an Agent that “executes tasks inside organizations.” Tencent's unique card is the consumer relationships and usage frequency connected to the WeChat ecosystem—WorkBuddy's entry from the C-end is essentially extending Tencent's strongest “high-frequency entry point” capability into work scenarios.

But the recent consolidation moves by all three also expose their respective anxieties. ByteDance folded the Feishu product team into the Doubao system, bidding farewell to the era when “collaborative office” played the leading role; Alibaba merged its three Agent tracks—QoderWork, Wukong, and MuleRun—into Qwen Office, concentrating its firepower in a catch-up effort; Tencent placed QClaw under WorkBuddy's management system—the smallest move, but equally clear in direction.

There is only one consensus: scattered experiments no longer make sense. There can be only one entry point.

The 2 billion yuan ARR and US$500 million annualized revenue both look on the surface like evidence that “AI office can make money.” Taken apart, they answer two different questions.

Doubao Work's ARR answers: are enterprises willing to pay for AI workflows? Reaching 2 billion yuan in three months suggests that within organizations Feishu already covers, AI capabilities are purchased as an extension of collaboration infrastructure, rather than as a new category requiring separate ROI justification. An iiMedia Research white paper shows that about half of enterprise payments use a hybrid “base subscription + excess usage” model—meaning the revenue ceiling depends on how deeply AI is used inside the organization, not on how many seats were sold.

WorkBuddy's DAU and token consumption answer a different question: can personal workflows generate high enough frequency that AI becomes an inseparable layer of daily operations? The 100 trillion daily average tokens give an affirmative answer.

But “high frequency” is not itself a business model. Between personal high frequency and enterprise-grade revenue lie four gates: subscription conversion, organizational procurement, compliance deployment, and IT integration. Tencent presents its revenue scale using “annualized from the last seven days,” which is effective for communication but leaves room on financial prudence—a month is the industry-recognized shortest reasonable annualization period, and the volatility of seven-day data is self-evident.

The two questions are not mutually exclusive, but their answers point to different organizational capabilities. Answering the first well depends on an enterprise sales system, customer success, and industry solutions; answering the second well depends on product experience, growth mechanisms, and ecosystem touchpoints. Answering both well is far harder than producing impressive numbers on any single path.

WorkBuddy's position in Tencent's financial report is worth noting.

In Q3 2026, Tencent's revenue was 207.336 billion yuan, up 8% year-on-year, while non-IFRS net profit fell 6% year-on-year; but cloud service revenue grew 35% year-on-year, and Tencent explicitly listed demand for WorkBuddy as one of the drivers. Against a backdrop of slowing overall growth, an AI office product entered the core financial narrative for the first time. This is itself a signal: WorkBuddy is no longer an “experimental product”—it carries Tencent's expectations of extending from consumer internet to enterprise services in the AI era.

The line from Pony Ma's internal speech in November—“land warfare has become naval warfare, and the army must become a navy”—is no exaggeration in this context. Tencent's absence from the top tier of AI-native applications is structural: the top three on QuestMobile's September AI-native App monthly-active ranking were Doubao with 419 million, Qwen with 158 million, and DeepSeek with 123 million, with no Tencent product making the list. WorkBuddy's 32.35 million office Agent monthly actives are one of the few assets Tencent has that can match up in the “work scenario.”

It must transform from “the fastest-running office Agent” into “an enterprise platform that can rein in revenue” in order to fill the thinnest piece of Tencent's AI map.

For the whole track, the two sets of data on the same day provide a clear observational framework: Doubao Work tests the sustainability of enterprise payments, while WorkBuddy tests the conversion efficiency of consumer-side frequency into organizational procurement. Neither test has reached its endgame.

What AI office ultimately competes for is not a new category, but the intersection of personal workflows and enterprise workflows. Here, DAU and ARR are merely snapshots of a given moment. What truly determines the landscape is who can enable AI to understand both an individual's operating habits and an organization's operating rules, and then connect the two into the same workflow.

The day of releasing data will pass. The battle for the entry point has only just entered the stage where it must prove “retention” rather than “traffic.”

暂无评论,快来发表第一条评论吧!No comments yet. Be the first!

📮 需求咨询📮 Inquiry