QuestMobile Q3 Report: Doubao Tops with 419 Million MAU, Tencent WorkBuddy Leads Office Agents

Since the start of this year, office Agents have gone through round after round of rapid iteration, and competition among the major tech companies now has more intuitive data comparisons.

On October 20, QuestMobile released its Q3 2026 AI application market report. In September, monthly active users in the AI office sector reached 142 million, up 50.6% year on year. Among them, Tencent WorkBuddy ranked first among office Agents with 32.35 million MAU, followed by Baidu Dazi and Qianwen Office.

Another list, of AI-native apps, presents a different landscape: Doubao firmly held first place with 419 million MAU, while Qianwen and DeepSeek ranked second and third with 158 million and 123 million respectively.

The battle for traffic among general-purpose AI assistants has gradually entered a stage of competition among the leaders, with the landscape stabilizing, and the major tech companies have now set their sights on office scenarios. Yet the existing lead may not carry over: Doubao is far ahead in the general-purpose AI assistant market, while Tencent was the first to seize the top spot among office Agents.

One track's landscape is taking shape, while another has just begun. Starting from these 19 apps, it is possible to see clearly how competition in AI applications is extending from phone screens to computer desktops, and what bets the major tech companies have each placed.

01. The top three AI assistants take shape, as major tech companies vie for the next entry point

QuestMobile's two lists—one of AI-native app MAU, the other of office Agent MAU—when viewed together, present two states of competition in the AI application market.

First, the familiar battlefield of the AI-native app list.

In September 2026, Doubao ranked first with 419 million MAU, while Qianwen and DeepSeek placed second and third with 158 million and 123 million respectively. The combined MAU of the three products accounted for more than 80% of the top ten total, opening a clear gap with the products behind them.

Among them, Doubao's lead is the most prominent, with MAU about 2.65 times that of second-place Qianwen. Fourth-ranked Tencent Yuanbao had 37.33 million MAU, no longer in the same league as the top three. This means that after more than a year of competing for users, the general-purpose AI assistant market has formed a relatively clear leading camp. For latecomers, replicating Doubao's user scale in the comprehensive AI assistant market is becoming increasingly difficult.

However, an emerging top-tier landscape does not mean the rankings are fixed. The list shows that Qianwen's MAU grew 5076.6% year on year, overtaking DeepSeek in one stroke, while the latter fell 15.2% year on year. Between the rise and the fall, two things are evident: first, user migration costs in this market are far lower than imagined—a single improvement in model capability or a wave of ad spending can rewrite the rankings; second, first-mover advantage has not translated into irreplaceable user mindshare.

Among them, Ant Afu ranked fifth with 34.79 million MAU, Doubao Aixue ranked seventh with 21.28 million, and Lovekey, which focuses on emotional companionship, ranked ninth with 8.98 million. The three products target scenarios such as health, education, and emotional companionship respectively. For the major tech companies, these products fill ecosystem gaps beyond the general entry points; for players without traffic entry points, this shows that choosing a high-frequency, rigid-demand scenario still offers a chance to gain a foothold outside the top tier—Lovekey being one example.

The variable truly worth watching lies in the second list.

On the office Agent MAU list, Tencent WorkBuddy ranked first with 32.353 million MAU, while Baidu Dazi and Alibaba Qianwen Office ranked second and third with 9.77 million and 7.905 million respectively. Products such as Tencent QClaw, ByteDance Doubao Work, and TRAE Work also entered the top ten.

Compared with the hundreds of millions of MAU boasted by general-purpose AI assistants, the user scale of leading office Agent products still shows a clear gap, but their growth rate cannot be ignored. In the third quarter of this year, WorkBuddy's MAU grew 647.1% quarter on quarter, and Baidu Dazi's grew 1567.5%. In just three months, the MAU of the two products grew about 6.5 times and 15.7 times respectively.

However, compared with general-purpose AI assistants, the landscape of this market is not yet stable. Judging by product development timelines, Qianwen Office's PC version launched in July, while Doubao Work's PC version and Baidu Dazi's personal edition launched in August. Many products have not been on the market long, and the current rankings reflect more each company's performance in the early scramble for users, which is not yet sufficient to judge the long-term competitive landscape.

Another noteworthy change is that the PC is becoming a new growth entry point for AI applications.

QuestMobile data shows that in September 2026, the PC client user scale of office Agents grew 512.2% year on year, close to the 518.4% increase on the app side. This means AI applications are undergoing an entry-point migration, and the PC surge is mainly driven by office scenarios.

From AI assistants on phones to office Agents on computers, competition among the major tech companies is extending into new scenarios. However, the user scale and traffic advantages accumulated in the past may not directly determine the outcome of the next round of competition.

02. ByteDance, Tencent, Alibaba, Baidu: four approaches vying for the desktop

The same company occupies vastly different positions on the two lists. ByteDance's Doubao is far ahead on mobile, while Doubao Work ranks only fifth on the office list; Tencent Yuanbao has fallen behind on mobile, yet WorkBuddy has taken first place on the office list.

The difference lies in the different needs they serve. The former is mainly aimed at individual users, while the latter, to truly enter enterprises, still needs to be integrated with workflows and internal enterprise systems. Over the past few months, the major tech companies have gone through multiple rounds of product iteration and business adjustments, with differing routes and priorities.

Let's start with ByteDance, which has the broadest product spread and the most frequent strategic adjustments.

In the two September rankings, ByteDance had five products listed: Doubao, Jimeng AI, Doubao Learning, Doubao Work, and TRAE Work.

More noteworthy than the number of products is ByteDance's reorganization of these businesses in recent months.

At the end of July, the Feishu product team was integrated with the Doubao product team; on August 24, ByteDance merged the TRAE and Coze teams as a whole into the Doubao system, and the next day released the standalone AI office product "Doubao Work." For a time, the outside world thought ByteDance was going to completely cut TRAE. But on October 9, TRAE Work and TRAE Code were merged and upgraded into the new version of TRAE, retaining the independent brand and putting office task execution and software development capabilities on the same platform. Doubao Work, meanwhile, continues to target broader office needs.

Image source / Doubao official website

From this, it appears ByteDance is actually doing two things: on one hand, using Doubao as the mass-market AI entry point, extending into scenarios such as education and office work; on the other hand, retaining professional products like TRAE to take on complex development needs that general assistants cannot fully cover. The advantage of this layout is that Doubao has a user scale other vendors cannot ignore, so ByteDance does not have to fight for AI users from scratch. But the difficulty is equally obvious: Doubao, with more than 400 million MAU, has not given ByteDance a leading position in the office Agent market.

Now look at Tencent, whose route is to concentrate its firepower, target office scenarios, and go all-in on WorkBuddy.

Tencent had three products listed: Tencent Yuanbao, WorkBuddy, and QClaw.

WorkBuddy's multi-platform MAU is already close to that of the Yuanbao app. Although the measurement bases differ, the growth gap is clear: the former grew 647.1% quarter over quarter, while the latter clearly declined. Although part of this is due to a low base, Tencent's investment in it is real. On the second-quarter earnings call, Tencent management said that after confirming WorkBuddy was taking off, it decisively shifted resources toward it while lowering the priority of some other new AI products in its portfolio. In addition, QClaw, which also made the list, announced in September that it would cease operations, further concentrating resources on WorkBuddy.

This choice is directly related to Tencent's entry-point structure. WeChat and WeCom already have ready-made, highly sticky work communication scenarios, and WorkBuddy can be embedded in collaboration workflows users already use, without needing to complete user education from scratch like a standalone product. By contrast, Yuanbao, facing pressure from Doubao, Qwen, and DeepSeek, has increasingly lower odds and cost-effectiveness in competing for the consumer entry point.

Judging from the current results, this approach is indeed working: WorkBuddy not only ranks first in MAU but has also opened a gap of more than three times over the products behind it. Before office scenarios enter commercialization, Tencent has at least secured a leading position.

Alibaba is the only player among the four to make the top three on both lists.

Alibaba had four products enter the two lists. On the AI-native app list, Qwen ranked second; on the office Agent list, Qwen Office ranked third, while Qoderwork and QwenPaw ranked seventh and ninth, respectively.

Qwen competes for the consumer general-purpose AI entry point, Qwen Office cuts into work scenarios, and products such as Qoderwork explore professional task execution. From this perspective, Alibaba's layout and approach are quite similar to ByteDance's.

The more striking figure is that Qwen grew 5076.6% year over year, surpassing DeepSeek in scale. However, on a quarter-over-quarter basis, Qwen fell 5.4%, and its catch-up momentum has already slowed; the high year-over-year growth rate is also partly due to a low early base.

Zhang Yi, chief analyst at iiMedia Research, told "AIX Finance" that Qwen's rapid growth is related to Alibaba's strong ecosystem traffic-driving capability and free strategy, and that Alibaba's long-term accumulation in the To B market also provides conditions for its extension into office Agents.

But Alibaba's problem lies here as well: it has Qwen's user scale and DingTalk's enterprise foundation, yet it has not yet shown an absolute advantage on any list. In addition, Alibaba previously integrated QoderWork, Wukong, and MuleRun under the "Qwen" brand, but how to define product positioning and coordinate resources across different teams is also a problem Alibaba needs to solve.

Baidu, meanwhile, is the fastest mover in this round of office Agent competition, but its foundation is relatively weak.

Baidu Dazi ranked second in office Agents with 9.77 million MAU, behind only Tencent's WorkBuddy, with quarter-over-quarter growth of 1567.5%, the fastest growth rate on the list. However, its app only launched in August, so a large part of this growth comes from the launch of the new form factor.

In the view of Ding Boran, an investor at Dewu Capital, whether Baidu Dazi can further enter the enterprise market remains uncertain. He believes Tencent has WeCom, Alibaba has DingTalk, and ByteDance has Feishu, and these products are already embedded in enterprises' daily communication and collaboration workflows. By comparison, Baidu lacks an enterprise collaboration entry point of comparable scale, making it more difficult for its Agent to penetrate deep into enterprises' internal business flows.

Of course, the lack of a proprietary enterprise collaboration platform does not mean Baidu can only rely on standalone products to acquire customers. Through third-party system integrations, enterprise customization, and specialized agents, Baidu likewise has opportunities to enter specific business scenarios. But whether Baidu Dazi's rapid growth can be sustained, and whether these users can be further converted into enterprise customers, remains to be verified.

Looking at the moves of the four companies together, Ding Boran judges that the four will ultimately end up looking very similar: "Each is pushing both to C and to B at the same time, and both will become focal points of competition," competing for the personal AI assistant entry point on one side while embedding Agents into enterprise collaboration platforms and business processes on the other. But he believes that having a massive base of general AI users is only the starting point of competition; because of WeCom, DingTalk, and Feishu, Tencent, Alibaba, and ByteDance have already secured half a ticket aboard the ship.

03. From Chatbot to Agent: What Will the Next Round of AI Application Competition Be About?

In September, the monthly active users of Doubao, Qwen, and DeepSeek all surpassed the 100 million mark, while WorkBuddy, ranked first among office Agents, had 32.353 million MAU. A clear scale gap still exists between the two product categories. This means the war over office scenarios has only just begun.

Zhang Yi believes that competition among AI applications is shifting from traffic entry points to task entry points, and the metrics used to measure product value will change accordingly. Rather than simply pursuing MAU, office Agents also need to be judged by task completion rates and final delivery outcomes.

Moreover, this competition is no longer confined to each company's own office ecosystem. Qwen Office has begun supporting migration of local historical data from WorkBuddy, and Doubao Work has also rolled out features for importing conversations, skills, and working memory from other AI tools; WorkBuddy, meanwhile, offers ways to connect to different collaboration platforms. As a result, accumulated work records, operating habits, and tool configurations are becoming new switching costs.

What the companies are fighting for now is the task entry point and the construction of data moats, but beyond these, the core will still come back to whether commercialization is possible.

Image source / pexels

The high MAU of general AI assistants is built on large-scale free access, and monetization has always been strained; office Agents, by contrast, confront efficiency value from the outset, with a much clearer path—individual subscriptions, enterprise procurement, and per-volume or per-task charging are all ready-made models.

In Ding Boran's view, enterprise office Agents may further evolve into "digital employees" in the future. Enterprises will no longer simply buy a piece of software, but instead have Agents take on specific tasks and pay based on usage or work outcomes.

But the threshold of the enterprise market is far higher than on the C-side: accuracy, data security, and the stability of task execution—failure to meet any one of these can lose a customer—and deployment and service costs are heavier. Ding Boran believes that enterprise office Agents still face an important obstacle: data security. Even if an Agent can complete tasks, enterprises may not be willing to hand over core business to it immediately.

In his view, the commercialization of office Agents still needs to go through a process from user activity and user retention to paid conversion and paid-user retention. The current rapid growth in MAU does not mean the product has established a stable business model.

And from the perspective of long-term competition, what determines the winner among office Agents may not necessarily be just underlying model capability.

Qwen's ability to overtake DeepSeek within a year through operations has already validated this logic. What is more likely to become a moat is still workflow integration capability and high-frequency scenarios. Zhang Yi listed three specific directions: workflow integration, the accumulation of organizational context, and the quality of vertical delivery.

The big tech companies each have their own advantages. Tencent can rely on communication entry points such as WeChat and WeCom; Alibaba and ByteDance have office ecosystems in DingTalk and Feishu, respectively; and Baidu is trying to connect different specialized agents through Dazi. They are all competing for a comprehensive entry point that covers more tasks.

In Zhang Yi's view, Tencent's self-developed large models are relatively a weak point, but its strategy is fairly clear; ByteDance's Doubao has the strongest traffic, and once its integration capability is filled in, it will only gain more advantage; Alibaba has deep B-side roots, and after Qwen's push on the C-side, it can also stabilize a leading position; only Baidu is "growing fast, but its ecosystem depth is relatively weak."

In addition, vertical players have not been eliminated either. The mobile rankings already provide a reference: beyond general assistants, segmented needs such as education, health, and emotional companionship have also produced products like Doubao Aixue, Ant Afu, and Lovekey.

The office Agent market may also replicate this kind of differentiation. General Agents handle and allocate tasks, while specialized Agents provide deeper capabilities in specific domains such as programming, design, finance, and sales. Ding Boran added that, for example, in terms of proprietary data and skills in niche industries, startups still have room.

The real suspense belongs to another type of player: companies with models but no scenarios. In a landscape where both traffic entry points and task entry points are gradually being occupied, how many positions are left for them?

Zhang Yi offered another angle. In his view, a standalone office Agent product with hundreds of millions of MAU may not necessarily emerge in the future; most users still use large-model products the way they used search engines in the past, and logically, players with strong model capabilities have a chance to have the last laugh.

This article comes from the WeChat public account "AIX Finance," author: Li Mengran, editor: Wei Jia, published by 36Kr with authorization.