GuShiio.com Wen| Blue Hole Business, author| Zhao Weiwei
The grayscale test was connected to DeepSeek’s WeChat AI search, and “Sorry, the service is busy, please try again later” frequently appeared.
WeChat may have been ill-prepared for the sudden influx of enthusiasm, but then WeChat gave a solution, which was to drain the flow of its own large-scale AI assistant “Yuanbao”. Starting from February 18, WeChat AI search users who have not yet been grey-tested will find that “Download Yuanbao (Experience DeepSeek R1)” appears on the page.
This model can be analogous to ByteDance’s Douyin for diverting Doubao APP, but ByteDance has built a super APP with a clear goal earlier than Tencent.
This move also means that Tencent Yuanbao’s market influence among C-end users will be greatly stimulated. According to QuestMobile, in the AIGC application in January this year, DeepSeek’s daily activity reached 22.92 million, the second-ranked bean bag was 15.86 million, while Tencent Yuanbao’s monthly activity was only 2.75 million.
Nowadays, users can not only use DeepSeek in the Yuanbao APP, but also use the latest in-depth thinking model T1 launched by Tencent’s Mixed Origin Model and the previously launched Tencent Mixed Origin Universal Model Turbo. This not only satisfies in-depth thinking, but also supports rapid output of answers.
The outside world has already held great expectations for this, because WeChat’s 1 billion + monthly activity data and high-quality content, coupled with DeepSeek’s online information integration capabilities, represent that WeChat search is striving to become the most powerful AI search form in China.
WeChat, which has always been very cautious in iterative upgrades of functions, has never made niche demands, but this time it has taken an extraordinary speed, which means that WeChat is adapting to the growth of a large user base and has begun to vigorously develop the AI To C market., fully participate in the competition, and DeepSeek’s open source model has become the biggest consensus in the Internet industry.
This year, Internet products have been connected to DeepSeek, essentially using AI technology to improve the intelligence level of C-end products and compete for market share, because it is not easy for the To C market to find a new growth path. For Tencent, social and games are the right and left hands. AI To C on social products is currently the most important A-side. The future B-side may be the combination of AI and games, driving the development and production of games on the B-side.
AI To C has become the general direction of Tencent’s AI model, so will Tencent’s AI To B road be far behind?
Make AI To C
This is not the first time WeChat has tested AI search in gray scale.
On May 30 last year, Tencent launched the AI assistant Yuanbao APP based on its self-developed mixed model; and five months later, WeChat users began to be grey-tested for the AI Q & A function. As long as they enter simple questions in the WeChat search box, the AI Q & A will be triggered.
Compare this year, DeepSeek launched an APP for ordinary users on January 15. Just a month later, WeChat launched a further “AI search” for grayscale testing. DeepSeek has brought great enthusiasm to WeChat.
The time difference between 5 months and 1 month proves that WeChat is accelerating AI To C and is more concerned about application scenarios rather than just adhering to self-developed models.
When will the official version of WeChat AI search be fully rolled out? At present, Tencent has not announced specific launch plans, only mentioning that it will decide the next step based on grayscale testing and user feedback.
The outside world obviously has expectations for this. Some independent developers have made suggestions for WeChat to upgrade the “WeChat Transfer Assistant” with file transfer function to “My Assistant” with broader AI functions.
In the tide of accessing DeepSeek, not only WeChat, Tencent as a whole has shown a high degree of participation.
Tencent’s only official official account issued an article on February 17, stating that Tencent’s Yuanbao, WeChat, ima, Tencent Document, QQ Browser, QQ Music and other Tencent products are connected to DeepSeek-R1 while accessing the Mixed Origin Model. model,”Welcome Experience”, and provides a link to a user guide for each business.
DeepSeek has undoubtedly had a significant impact on Tencent’s big model plan.
After DeepSeek connected to Tencent Yuanbao APP, Tencent recently launched the Mixed Origin Deep Thinking Model T1 (Thinker) to provide a testing experience. Users can choose different models to compare and use them. Mixed Origin T1 is similar to DeepSeek’s open source R1 model. Both of them perform well on complex issues through inference models, but they are strong in supporting real-time search of public account articles.
WeChat grayscale tested DeepSeek, and after Yuanbao connected to DeepSeek, it quickly began grayscale testing the mixed color T1. It can only be said that Tencent, which was not slow in the field of large models before, has stepped up its pace of catching up in 2025 and showed the most positive attitude: “Make AI To C vigorously.”
And a bigger wave has arrived. A large number of Internet companies are using DeepSeek to attract user experience. The nano-AI search under the 360 Group, nailed AI assistants, and multi-dimensional tables in Byte’s flying books have all realized access to DeepSeek. Access calls, DeepSeek is a lesson that large companies cannot miss.
For ordinary users, who is the one that satisfies them most?
For WeChat users, future AI search habits will definitely be more free and vertical, and it will be easier to achieve a context-based service closed loop.
If you search for “Make a May Day Shanghai travel guide and plan, requiring you to avoid crowded places,” WeChat will make itinerary arrangements based on the time and even suggestions from friends ‘circles. When ticketing is involved, you can directly jump to 12306 and other ticket purchasing Mini programs, thereby realizing price comparison and ticket purchase, and automatically recommending hotels and itinerary arrangements, etc., saving users time.
Of course, WeChat AI search emphasizes that its AI search only integrates public accounts and other public information on the Internet, and will not use users ‘personal information and related private information.
For Tencent, the value of WeChat AI search contribution will gradually emerge in the future. 600 million searches per day will provide a greater increase in advertising business revenue.
Tencent mentioned at its third quarter of 2024 earnings report that WeChat search utilizes large-scale language model capabilities to help users understand complex queries and content, improve the relevance of search results, increase commercial queries, increase click-through rates, and “search revenue has doubled year-on-year.” This is the first time that Tencent separately disclosed Souyisou’s revenue data last year.
This is only the result of the support of the Mixed Origin Model. So how much will WeChat AI search connected to DeepSeek contribute to Tencent in 2025?
To B cooperation is the key?
It can be said that among the businesses that Tencent contributes the most revenue, the only thing missing is games that have not yet connected to DeepSeek.
As Tencent’s “biological son”, the Hunyuan Model has also been widely used in various businesses last year, including Tencent Yuanbao, Weixin Official Accounts, Mini programs, WeChat reading, Tencent advertising, Tencent games, Tencent Meeting and other nearly 700 internal businesses and scenarios, all of which have applied the capabilities of the Hunyuan Model.
Why is Hunyuan connected to Tencent games, but DeepSeek has not yet?
In fact, the AI capabilities of the big model are more beneficial to Tencent in its advertising business. They are limited in use in the game business. The mixed model mainly helps the game business provide customer service. Tencent President Liu Chiping admitted at the first quarter of 2024 earnings report. AI is not yet fully competent for the important task of content creation in games, and requires several iterations to reach the level suitable for game production.
One year has passed. When will the Mixed Origin Model truly drive Tencent’s game business? The answer may be this year.
Tencent open-source two models in November last year, one is the big language model “Mixed Origin Large”, and the other is the 3D image generation model “Mixed Origin 3D”. “Mixed 3D” aims to help developers create the production of 3D content and save production cycles. At the time, Tencent disclosed that “it plans to start trialing 3D large model generation assets in game products in 2025.”
This year, not only will the mixed 3D model test the water in the game business, but also DeepSeek’s efforts in the multimodal business.
According to online research records on DeepSeek, DeepSeek will launch two main models before June this year. One is DeepSeek-R2 and the other is DeepSeek-V4. It is an upgrade to the current open source models R1 and V3, including innovation in multimodal architecture, and “adopts more radical innovative solutions.”
In other words, this year DeepSeek will launch a multimodal model, and Tencent will also use the multimodal model in game production.So, the highlight of the technology industry this year is not only that WeChat connects to DeepSeek for AI search, but more importantly, will Tencent Games connect to DeepSeek for game production?
To answer this question, Tencent’s game business group and technology engineering business group need to cooperate together.
Compared with breaking AI To C through DeepSeek, Tencent’s use of multimodal large models to improve game production is a more difficult topic. This is also a longer-term plan that needs to break through many key technical bottlenecks. Prior to this, AI enriches game content and gameplay, which may be a faster development direction.
More importantly, it involves Tencent’s huge internal ecosystem and different divisions of labor and cooperation among teams.
The games business group, WeChat business group, technology engineering business group, cloud and smart industry business group all have their own AI teams responsible for different fields. Just in January this year, Yuanbao products were adjusted from the technical engineering business group to the cloud and smart industry business group, which means that Tencent’s division of labor at the model level and application level is clearer, and the cloud and smart industry business group focuses on application implementation and commercialization.
Focusing on making models improves model capabilities and focusing on the market for applications is one of the important reasons why ByteDance can develop rapidly in AI capabilities in 2024. At the end of 2024, Alibaba’s “Tongyi” team was split from Alibaba Cloud and merged into the Alibaba Intelligent Information Business Group. The same is true: the underlying technology and application products are separated and focused on their respective challenges.
In 2025, Tencent followed Ali and Byte to set up a business formation and truly began to meet the test of the AI To C era.
In the future, how to combine King’s Glory with the Mixed Origin Model or DeepSeek multimodal will be Tencent’s real imagination space in the future. The connection of Tencent products such as WeChat and Yuanbao to DeepSeek represents the enhancement of Tencent’s To C power, while the connection of game products to DeepSeek means that Tencent’s To B capabilities are powerful.
It is not enough to make a final conclusion on WeChat search now. The more important reality is that WeChat search will evolve from an “information integrator” in the past to an “intelligent service hub.” This should be based on the closed-loop search of ecological components such as serial public accounts, Mini programs, and video numbers, and further achieve a leap in AI search capabilities.
Four months later, when DeepSeek releases its multimodal model, the Internet technology industry may be another huge earthquake.
