WeChat integrates with DeepSeek, and Tencent Yuanbao plays the "divide and conquer" strategy.

Text｜Wang Yuchan

Editor｜Qiao Qian

In recent days, the new trend among Internet companies is to connect their own products to DeepSeek. Among this trend, the participation of WeChat is particularly remarkable.

On February 15, WeChat Search began a gray-scale test of the "AI Search" function, integrating the DeepSeek-R1 model into the WeChat search function. Users in the gray-scale test can click on "AI Search" in the WeChat search box to freely use the full-version model of DeepSeek-R1.

As a super application with 1.382 billion monthly active users (combined WeChat and WeChat for international users), the cooperation between WeChat and DeepSeek is of great significance. Moreover, Tencent, which has always been cautious with the style of "daring to be the last to act", has joined this trend, which is also somewhat unexpected.

Three days after the gray-scale test began, the cooperation between the two sides is still deepening. On February 18, 36Kr learned that Tencent internally urgently coordinated and deployed the invocation of "Yuanbao" to support WeChat Search, providing users with more entrances and choices. Tencent Yuanbao is an AI application based on Tencent's Hunyuan large model.

Tencent stated, "Recently, WeChat Search has been conducting a gray-scale test of AI Search, and users' enthusiasm for using the DeepSeek model has far exceeded expectations. Starting from today, for users who have not yet been included in the gray-scale test of AI Search, by clicking on the search bar at the top of the WeChat dialog box and entering WeChat Search, they will have the opportunity to see 'Download Yuanbao (Experience DeepSeek-R1)' on the page."

Through Tencent Yuanbao, users can freely use the full-version DeepSeek-R1, DeepSeek-V3, Tencent Hunyuan Deep Thinking Model T1, and Tencent Hunyuan General Model Turbo, which not only supports deep thinking but also supports quick output of answers.

In addition, after obtaining the answers from DeepSeek within WeChat and the Yuanbao App, users can also support sharing the question-and-answer process with WeChat friends with one click, and display it in the form of a card in the dialog box.

DeepSeek is a Chinese company specializing in artificial intelligence large model technology, dedicated to researching and developing and providing advanced natural language processing (NLP), computer vision (CV), and multi-modal AI technologies. Since the company launched the open-source large model DeepSeek-R1 in January this year, it has been in the spotlight. The attention of WeChat users to this new technology even exceeds that of WeChat's important new functions - WeChat Index shows that during the Spring Festival, the search popularity of "DeepSeek" is more than five times that of "WeChat Small Blue Packet".

For Tencent Yuanbao, which is slightly slower in the C-end competition, connecting to DeepSeek is far from being as simple as "enriching the user experience".

What impact will WeChat's connection to DeepSeek bring?

The cooperation between Tencent and DeepSeek is interpreted by the outside world as "moving 1.3 billion users into DeepSeek", "enlarging the ecosystem" or "a signal of change". Considering the influence of the national social software, it is reasonable for everyone to make such judgments.

However, it needs to be clear that the current gray-scale test function within WeChat seems far from mature - using AI Search within WeChat not only cannot understand the context, does not support multi-round conversations, does not support uploading various file contents to assist in questioning and searching, and even does not support viewing historical question-and-answer records.

In a word, for those who really need to use DeepSeek, it is much better to directly download a DeepSeek from the app store than to use "AI Search" in WeChat.

Wei Xi, a tech author who focuses on the Internet and AI, joked that the connection of DeepSeek to WeChat Search seems to be as hasty as "sticking DeepSeek to WeChat Search with transparent tape". The only two features that can reflect the presence of product managers are: one is that the search results support sharing, and the other is that there are "recommended questions" after the output results.

He believes that the biggest possibility for WeChat to launch an immature function so quickly is that the team wants to quietly and quickly train to obtain feedback, so it designed it to be "easy to access but with a shallow experience". However, the status of WeChat itself as a "super App" and the recent high popularity of DeepSeek have caused the outside world to amplify the attention to the new function and overestimate its short-term impact.

However, there is no need to be pessimistic about the new function. Because in the long term, the cooperation between the two is still of great significance.

"The core of DeepSeek as an independent App is 'deep thinking' + 'connected search'. " Wei Xi said. "There is no ability gap in 'deep thinking' for any App that connects to DeepSeek, and the core difference in the experience layer lies in 'connected search'. " Therefore, for the same question, the answers are different for Zhihu connecting to DeepSeek, Xueqiu connecting to DeepSeek, Xiaohongshu connecting to DeepSeek, and WeChat connecting to DeepSeek due to different data sources.

WeChat's connection to DeepSeek further breaks the information island of APPs and brings the valuable assets accumulated in the public account ecosystem. In addition, the social attribute of WeChat itself is also more conducive to DeepSeek cultivating users' habit of conversing with AI robots here.

"Let's simply have a wild imagination - if this AI conversation robot is directly open to the personal address book, allowing different AI agents to become your chat friends at any time, the experience increment will be an order of magnitude higher than the current makeshift product. " Wei Xi said.

As for the imagination of "AI helping users make consumption choices" that is closer to money, it seems too early to talk about it now.

On February 18, a blogger claimed that when Tencent's intelligent assistant Yuanbao was using DeepSeek-R1, the answers seemed to contain suspected advertising information. Zhang Jun, Tencent's public relations director, responded in the comment area: "Theoretically, it shouldn't be. We have asked the technical team to check it." Subsequently, Zhang Jun further explained that the reason for the problem is that the answer referred to the pit-avoidance content of the "58 Daojia Home Cleaning, Babysitting, Maternity Matron, and Moving" public account, and this content may come from the connected search, resulting in the accidental embedding of advertising information.

Should big companies continue to develop their own large models?

Tencent's acceptance of DeepSeek makes people wonder about the group's next plan for its own Hunyuan large model. After a super product has emerged and quickly occupied the market, how should big companies' self-developed large models position themselves?

With the sudden emergence of DeepSeek, several giant companies that have already deployed AI large models have quickly abandoned "protectionism" - Baidu, Alibaba Cloud, Huawei, and Tencent have successively connected to DeepSeek, while ByteDance's Doubao is still observing.

ByteDance's hesitation is justified. Before the appearance of DeepSeek, ByteDance's AI assistant Doubao was the AI conversation application with the largest number of monthly active users in China, followed by Kimi and Wen Xiaoyan. While Ali Tongyi and Tencent Yuanbao ranked at the bottom.

DeepSeek has led to a sharp increase in the demand for C-end users to experience AI. Several big companies that did not perform well in the C-end naturally accepted DeepSeek, while Doubao, which originally had the opportunity to become the first, obviously does not want to give up easily.

At the recent 2025 ByteDance All-Hands Meeting, ByteDance CEO Liang Rubo said: "One of the innovation points of DeepSeek-R1, the long-chain thinking mode, is not the industry's first. After OpenAI released the long-chain thinking model in September 2024 and it became an industry hotspot, ByteDance realized the significant technological change, but 'did not feel the need to immediately replicate it'... Looking back now, if we had competed for major issues at the beginning, we would have had the opportunity to achieve it earlier."

A report from Zheshang Securities stated that ByteDance's capital expenditure in the AI field in 2024 reached 80 billion yuan, close to the total of Baidu, Alibaba, and Tencent (about 100 billion yuan). Liang Rubo said that in 2025, ByteDance's AI business will pursue the "intelligence ceiling", explore new interactions, and strengthen the scale effect. It can be predicted that Doubao, with more than 70 million monthly active users, will continue to compete with DeepSeek in the C-end.

Tencent Yuanbao, which is temporarily one step behind in the C-end, connecting to DeepSeek at this time can also be regarded as a risky move of "driving a tiger to swallow a wolf". At the beginning of this year's annual meeting, Ma Huateng said that Tencent will continue to invest resources in computing power reserves, and hopes that all BGs can embrace the productization and landing scenarios of large models. Currently, products including WeChat, QQ, Input Method, and Browser will launch AI agents.

Can giant companies break the innovation curse and surpass DeepSeek in the next competition? Wei Xi said that Tencent's complex and huge product ecosystem requires the multi-modal capabilities of Hunyuan (such as generating advertising creativity), and multi-modality is precisely the shortcoming of DeepSeek at this stage.

"So I cautiously believe that Tencent will not reduce the investment in its self-developed base model, but rather urge them to come up with more 'hard goods' with leading indicators with higher efficiency under the support of abundant resources." Wei Xi said.