From being 'extremely conservative' to becoming the first chat application to integrate DeepSeek, WeChat's noticeable change indicates that its team recognizes the immense potential of the reasoning model represented by DeepSeek R1 for application on the WeChat platform and has decided to quickly engage and lead this wave of change. As an important communication platform today, WeChat has become an indispensable part of the process to 'truly popularize AI.'

Since its launch in 2011, WeChat has been known for its founder Zhang Xiaolong's "restrained" approach, not changing WeChat itself for the sake of a "trend." Even today, the experience of scrolling through Moments is not fundamentally different from when it first emerged ten years ago.

"Zhang Xiaolong feels that one of the aspects he is most satisfied with regarding this feature is that it has been stable and running for ten years with almost no room for improvement after its release." This is how Zhang Peng, founder of Geek Park, summarized WeChat's product logic after a dialogue with Zhang Xiaolong. This has not changed since WeChat became a true "national social app."

After the release of DeepSeek R1, these tags were suddenly torn off like masks by the WeChat team: On the evening of February 15, users noticed that an "AI search" feature integrating DeepSeek R1 had appeared in their WeChat search—less than a month after DeepSeek was officially released and open-sourced on January 20.

For users receiving this update, tapping into the search bar at the top of the WeChat homepage reveals an "AI search" button below the search history. Clicking on it shows two response model capability options provided by this feature. The first is called "Quick Response," and the second is the "Deep Thinking" mode powered by the open-sourced DeepSeek R1.

It is worth mentioning that although this is only a grayscale update received by a portion of users, the speed of this update is much faster compared to the recent gray-scale feature updates like Callkit on WeChat, based on the range of the PUSH and the distribution of users who received the update. From the initial leak to the large-scale appearance of the first users who received the PUSH, it took less than six hours, completely overturning the conservative impression of WeChat's feature updates over the past few years.

This update is not bound to the version number of the Software itself, so there is no need to update the WeChat version from the App Store or other application stores. The author personally discovered this entry after manually clearing the WeChat background on the phone and restarting it.

In other words, if your WeChat has not yet received this update, there is no need to rush— the DeepSeek R1 entry in your WeChat is actually hidden in the code, and the testing qualification may be coming soon.

From the open-source and acknowledgment statement attached to the feature, it can be seen that the DeepSeek R1 built into WeChat is based on the open-source version, but it does not explicitly mention the model size used, whether it is the 671B 'full-blooded' R1 version.

It is well known that the WeChat content ecosystem, including public accounts and video accounts, has long been an 'island' independent of major Search Engines. This situation has not improved even after the emergence of internet-connected language models like ChatGPT4o, so when testing the built-in R1 in WeChat, I am most excited about whether it can finally freely soar in the ocean of WeChat's content platform using the capabilities of large models.

Unfortunately, the answer is negative in the current practical tests.

For example, I tried feeding it the WeChat official account's article link, but it was unable to retrieve relevant information from the WeChat platform, only being able to search for related content on the Internet based on fields from the link, and sometimes it couldn't even recognize that it was a link from a WeChat official account article.

In other questions related to WeChat official accounts, although AI search can provide relatively accurate responses based on article searches and R1's reasoning capabilities, the source of its content is closer to results derived from the distribution of WeChat official accounts on other platforms, rather than direct references to results from the WeChat content platform itself.

Currently, the built-in R1 in WeChat has not enhanced the generation for content retrieval within the WeChat platform (RAG), optimizing the output results.

This is not only an obvious shortcoming of the current beta version of WeChat's AI Search but also one of the significant directions for the integration of AI features in WeChat in the short term.

Overall, the current WeChat AI search experience is also very lightweight. It not only does not support continuous conversations but also does not allow the uploading of various file contents to assist in questioning and searching. Moreover, even after exiting the chat interface, the current dialogue memory content will be directly destroyed and not retained.

As the 'national app', WeChat integrated an entry within less than a month after the official release and open-source of DeepSeek R1, which is undoubtedly an exciting development.

The excitement comes from the fact that what WeChat is doing can only be achieved by a few manufacturers.

After the conversational generative model has a mobile app, seizing the entry point on the Smart Phone desktop is something that many AI applications, including ChatGPT and Perplexity, are doing by setting it as the default voice assistant, trying to take 'control of the entry' from mobile brands.

However, for a relatively closed platform like WeChat, it seems that only the WeChat team itself can rely on the existing vast Chinese content ecosystem to do and do well in this regard.

From being 'extremely conservative' to now being among the first chat applications to join DeepSeek, such a significant change has only one reasonable explanation: the WeChat team has recognized the tremendous potential of the reasoning model represented by DeepSeek R1 for application on the WeChat platform and has decided to quickly engage to become the leader of this wave of change, ensuring that the user experience on WeChat does not fall behind other competitors.

This is not the first time the WeChat team has taken action in the field of AI large models. The WeChat input method added the 'one-click AI Q&A' function in June last year, allowing users to receive content responses from language large models within the WeChat input method.

However, at that time, this function was based on Tencent's own Hunyuan AI large model and could not serve as a text generation tool.

In terms of experience, WeChat Input Method is suitable for "flash of inspiration" questions in various chats. The AI Search of the WeChat Ontology is likely to focus on the existing content ecosystem within WeChat, deeply mining application scenarios with the help of user chat content and platforms like public accounts.

Such a "distributed" experience of AI capabilities is somewhat similar to the product idea of Apple Intelligence: in the capabilities released by Apple last year, Apple did not introduce a revolutionary model to shock the world, but chose to embed the capabilities of existing models like ChatGPT in various corners of the mobile ecosystem including notes, photos, and input methods.

Superficially, Apple Intelligence seems less intelligent compared to other manufacturers like Google and OPPO, which are also working on mobile AI applications. However, what Apple is actually doing is allowing AI to join users' lives as quietly as possible while using its capabilities to bring changes to more usage scenarios.

From this perspective, although Tencent missed one of the most important entry points in the mobile Internet era without a mobile operating system, WeChat, as an important communication platform today, has become an indispensable part of the process to "truly popularize AI."

In China, from centenarians to children who have just learned to use a Smart Phone, everyone has acquired at least a basic understanding of how to use WeChat. Among these functions, there are also key aspects that are most suitable for AI capabilities to further do "smoothing" and reduce the learning costs of AI.

For generative AI, which has already passed its explosive point, the true future that WeChat AI deserves our expectations lies in exploring the popularization of AI applications, allowing AI capabilities to "produce qualitative changes" through more users' long-term use.

Even now we can conclude: the AI capabilities in WeChat may not be the most exciting one, but it has the greatest opportunity to truly "change the world."

Editor/new