On January 20th, The Wall Street Journal reported that Baidu’s Wenxin Assistant had soared past 200 million MAU (Monthly Active Users). Just two days later, on the 22nd, Wenxin unveiled a brand logo upgrade. Not long before this, QuestMobile reported that ByteDance’s Doubao had hit 172 million MAU, while Alibaba’s Qianwen APP had also surpassed the 100 million MAU mark. It’s safe to say that the “Three-Kingdom Face-Off” competitive landscape among domestic AI applications has largely taken shape, with Wenxin, Doubao, and Qianwen emerging as the “super gateways” for consumer-side AI. This scenario is evident not just in the statistical data but also in user perception and daily experiences, and this competitive dynamic is likely to persist for years to come.

Looking back at the development trajectory of China’s AI applications over the past few years, the most significant shift, in my view, is the transition from a pre-2025 focus on isolated technical indicators of models—such as parameter scale, benchmark rankings, and specific vertical functions (e.g., programming)—to a growing emphasis, starting from 2025, on the comprehensive application capabilities of large models, or the “application ecosystem,” by both professionals and ordinary users. The core question has shifted from (or is no longer solely) “How powerful is your base large model?” to “What can you actually do?” It’s no wonder some believe that 2026 will mark the true dawn of AI Agents—where an Agent is an artificial intelligence capable of executing a series of complex instructions, with practicality as the key battleground.

Therefore, drawing a conclusion is straightforward: the “super gateway” of the next-generation internet will be AI super applications boasting rich functionalities. These applications will transcend the current “chatbot” category, achieving a so-called “closed-loop for getting things done.” No matter what challenges users face in work or life, they can assign tasks to AI applications in natural language, and the AI will leverage various internal functions and external applications to complete the tasks, either instantly or over time. This transformation is already quietly underway, and if you haven’t experienced it yet, you soon will.

So, the question arises: Why are Baidu Wenxin, Doubao, and Qianwen the “Big Three” of domestic AI super gateways? To some extent, this reflects the natural evolution inherited from the mobile internet era. Doubao capitalizes on the massive traffic and content ecosystem of ByteDance’s apps like Douyin and Toutiao, combined with ByteDance’s consistently efficient product iteration and marketing prowess, quickly amassing a large user base. From my observations, a notably high proportion of light AI users favor Doubao. On the other hand, Qianwen leverages Alibaba Cloud’s technical prowess and is deeply integrated with Alibaba’s e-commerce businesses, such as Taobao/Tmall, Taobao Deals, and Fliggy, striving to achieve the “shortest path closed-loop” from conversation to transaction. This was also a major highlight of Qianwen APP’s recent upgrade.

What about Baidu Wenxin’s strengths? I believe it shares some similarities with its competitors but also boasts unique advantages. First, Baidu’s own application ecosystem remains rich and well-established. Baidu Wenxin Assistant has already integrated core content like Baidu Search, Baidu Netdisk, and Baidu Maps. Especially search, its importance will only grow in the AI era. Google’s market capitalization has repeatedly hit new highs over the past year, thanks to the seamless integration of its Gemini large model with its core search business. AI will not “make search obsolete” but rather leverage it to play a more profound role, shifting from “providing information” to “providing services.” If you, like me, are a dual user of Gemini and Wenxin, you’ll likely agree with my observations.

Second, compared to its rivals, Baidu Wenxin’s AI ecosystem places a particular emphasis on “openness.” It early on opened up MCP (Model Context Protocol) to a vast array of third-party applications, enabling it to invoke and deliver third-party services through its own interface. At this moment, Baidu Wenxin Assistant can directly tap into high-quality third-party services across multiple domains through MCP: e-commerce like JD and Taobao, local life services like Meituan, Dianping, and Maoyan, travel services like Ctrip and Tongcheng, finance like Yingmi Fund, and automotive like Autohome and Yiche...

For instance, just now I asked Baidu Wenxin Assistant, “What movie theaters are nearby?” and “What movies are currently showing?” It immediately invoked Meituan’s services, providing me with links to nearby theaters and showtimes. I then inquired, “How can I buy tickets for Avatar 2?” and it promptly offered me ticket-purchasing links from two major platforms, Maoyan and Taobao Tickets. Before purchasing, I also asked it to “list the price table for Avatar 2 at different theaters and hall types nearby,” and it even remembered to inform me of the theater’s refund and rescheduling policies. This was the first time I felt that using an AI assistant to find and buy movie tickets was more convenient and cost-effective than directly opening the relevant APP.

It can be said that Wenxin Assistant seamlessly integrates these services to provide users with more detailed information and a broader range of choices, while also offering service providers an additional avenue to reach users. This approach to building an open ecosystem aligns with the philosophy of search engines: no search platform can generate all content on its own; the richness of search content hinges on third-party content supply. In the mobile internet era, a trend emerged that was unfavorable to search engines: major content platforms “closed themselves off,” keeping content within their own platforms and inaccessible to search engines, creating “information silos.” This may benefit the content platforms themselves but is clearly detrimental to the overall user experience of the internet.

An AI application ecosystem based on MCP may hold the key to solving these issues. The core competitiveness of AI assistants lies in providing both information and connecting to services. As long as they can manage their relationships with third-party service providers and foster sufficient business growth, they can achieve what traditional search engines cannot—building a fully open ecosystem. From this perspective, Wenxin’s significance to Baidu may even surpass Doubao’s to ByteDance or Qianwen’s to Alibaba!

Of course, the vision of an AI super gateway has only been partially realized thus far. For example, users will find that in most cases, AI applications can only provide links to products or services and cannot directly complete transactions. If a request involves multiple different service providers, the complexity increases significantly. The process of converting information integrated by AI assistants into actual actions is often not entirely seamless. The long-term vision of an AI super gateway is to play a role akin to an “operating system,” seamlessly coordinating all information and services—what can be achieved now is probably only equivalent to the Windows 1.0 operating system of its time, with immense potential but a long road ahead before full realization.

The competition remains rife with uncertainties. I believe that in 2026 and even the next 2-3 years, the competitive landscape of the three major super gateways will hinge on the following three factors: First, progress in base large models remains crucial, but the focus may shift to the multimodal domain, especially the generation quality of video and audio, which will directly shape the user experience. Second, the integration between AI applications and existing super APPs within their own ecosystems must become tighter, including the integration of Wenxin with Baidu, Doubao with Douyin/Toutiao, and Qianwen with Taobao. Whoever provides a superior integration experience will be able to convert users within their own ecosystem into heavy AI users more rapidly. Third, the development of a third-party application ecosystem based on MCP, as no single major company’s “first-party services” can meet all user needs, and third-party services will ultimately become the deciding factor.

In summary, the competitive barrier in AI applications is shifting from “how many users you have” to “how many reliable services you can integrate.” The competition among the three is not just about who possesses stronger technical capabilities or greater marketing investment but also about who can integrate a sufficiently rich service ecosystem and “reconstruct” the internet with an open mindset.

Who will ultimately emerge victorious in this round may take a considerable amount of time to be revealed, but it will certainly be a spectacle worth watching.