On June 8, 2026, the WeChat Developer Platform announced that WeChat AI has entered the internal testing phase. This AI assistant, integrated within the WeChat ecosystem, allows users to directly invoke, access, and operate mini-programs through natural language conversation. The open platform provides two access modes: the automatic mode allows authorized platforms to read mini-program source code, enabling AI to directly operate the page without additional development; the development mode allows developers to build skills independently, which can be called by AI after review. The terms of service also state that "WeChat AI" may be a temporary name, the final naming has not been determined, and the access is optional and does not affect the normal operation of existing mini-programs.
This is the first time that WeChat has opened its mini-program ecosystem to AI at the conversational entry level. The background at this time is: Tencent's self-developed Mixian large model has entered the top tier in domestic public benchmark tests, and the Yuanbao App experienced explosive growth during the 2026 Spring Festival red envelope period, surpassing 100 million monthly active users. The internal testing of WeChat AI is the latest step in Tencent's AI journey from technology reserves and independent product verification to super application delivery. The automatic mode requires developers to provide source code; how many developers this low-threshold path can attract and what ecological interests and conflicts it may encounter are the questions to be answered during the internal test phase.
Opening of the Conversational Layer in the Mini-Program Ecosystem
The two access modes of WeChat AI target completely different developer groups.
The design logic of the automatic mode is straightforward: the authorized platform reads the mini-program source code during the review, the platform automatically analyzes the page structure, allowing AI to operate the page directly without additional development. A small game team of only two or three people does not need to equip an AI engineer, nor do they need to understand the Agent protocol; just by selecting authorization, their ordering mini-program or tool application can be called by WeChat AI.
According to data disclosed in the WeChat public class in January 2026, the WeChat mini-game ecosystem has gathered over 400,000 developers, 80% of whom are small teams with fewer than 30 people. In 2025, the overall daily active users exceeded 100 million, with monthly active users surpassing 500 million. This supply-side scale is a unique moat for WeChat AI. Byte's Doubao or Ali's Tongyi Qianwen can create an independent app and open an API, but they do not have a mini-program ecosystem with daily active users exceeding 100 million that can be directly accessed. The automatic mode of WeChat AI essentially exchanges technical conveniences for scalable access, allowing the vast majority of the 400,000 developers to board this train at zero cost.
The development mode retains customization space for service providers with complex business logic. Developers can build skills based on their business characteristics, which can be called by WeChat AI after platform evaluation and review. Both modes can be enabled simultaneously and are not mutually exclusive.
The phrases "name undecided" and "optional behavior" indicate that the WeChat team still has reservations about product positioning. The main task in the internal testing phase is to streamline the technical chain and observe developer reactions. However, the automatic mode has already touched on a sensitive point: source code authorization. Some developers have expressed concerns in the WeChat open community, focusing on a few core issues—how to ensure the security of code assets after the platform reads the source code, whether AI's direct operation of the page will render existing buried points and advertising display logic ineffective, and how to delineate liability if AI misoperation causes user losses. Currently, there are no publicly disclosed detailed regulations regarding these issues.
After Being Ranked Second in Basic Capabilities Domestically, Mixian Chooses to Go Deeper
What WeChat AI needs is not just a model that can chat; it needs an Agent foundation that can understand page structures and accurately execute operation commands. This foundation is Tencent's Mixian large model.
In March 2025, the Chinese large model evaluation benchmark SuperCLUE released a report, ranking Tencent's Mixian flagship version second in the domestic basic model standings, second only to Byte's Doubao; however, it ranked first in application capability, leading in subcategories such as text understanding and generation, instruction compliance, and Agent capability. Science.net pointed out while quoting this report that Mixian's performance in the "practicality" dimension exceeded its ranking in basic capability. At the same time, Mixian Turbo S made its first appearance in the international evaluation Chatbot Arena's global Top 15.
Mixian's version iteration maintains a quarterly rhythm. In April 2025, the hunyuan-turbo was updated, and in July, the flagship version TurboS was launched, enhancing its cognitive abilities. In April 2026, the Hy3 preview version was released, with the official claim that reasoning efficiency improved by 40%. According to Tencent Cloud product documentation, the old version HY 2.0 is scheduled to be discontinued starting June 26, 2026.
This pace is much slower than that of Byte and Ali. In the past year, Byte's Doubao and Ali's Tongyi Qianwen maintained a near "weekly update" model release frequency, while Mixian stabilized at one major version update per quarter. Tencent's management has previously made public statements about "slow work produces fine products." The technical explanation is that the Agent era requires far higher stability and low latency than the conversation era; frequent switching of underlying models would make it impossible for developers to adapt their engineering. The scenarios WeChat AI needs to invoke include ordering, payment, and appointment operations involving funds and sensitive information, where the determinism of model output is much more important than creativity.
In terms of resource investment, Tencent President Liu Chiping revealed in the 2025 earnings communication meeting that in 2025, AI new product research and development investment would be 18 billion yuan, and in 2026 this investment would at least double. The content reported by The Paper showed that Liu Chiping also stated that the next core plan is to create a dedicated AI intelligence body within WeChat, linking mini-programs, social networking, and payment throughout the entire chain. While the investment doubles, the version rhythm does not accelerate, indicating that more funds are flowing into the reconstruction of infrastructure and improvement of data quality rather than seizing release windows.
Mixian's lead in application capability resonates with WeChat AI's scenario needs. A basic model with a higher ranking but weaker Agent capability may not be as useful in WeChat AI's scenarios as Mixian. Tencent has chosen a path of not chasing parameter competition and focusing on practical dimensions, which began to show its logical coherence during the internal testing of WeChat AI.
Daily Active Users Surpassing 50 Million During Spring Festival, What Next?
Before the internal testing of WeChat AI, the C-end validation task of Tencent AI was undertaken by the Yuanbao App.
The growth curve of Yuanbao shows a clear pulsing characteristic. According to QuestMobile monitoring data reported by China National Radio, in January 2025, Yuanbao ranked 12th in monthly active users in the industry, rising to 3rd place by December 2025, second only to Doubao (MAU 226 million) and DeepSeek (MAU 135 million), with an annual compounded growth rate of 27.8%.
During the Spring Festival of 2026, Yuanbao experienced an explosion. Data disclosed by Tencent officials shows that Yuanbao's DAU peak exceeded 50 million, reaching 40.54 million on New Year's Eve, with MAU reaching 114 million. Shanghai Securities News reported that this growth mainly came from the social chain brought about by red envelope activities.
However, the data quickly declined after the Spring Festival. QuestMobile monitored that in April 2026, Yuanbao's normal DAU was about 9 million, while Doubao's DAU was about 14 million and Qianwen about 30 million. The peak-to-valley difference approached 5 times, indicating a significant pulsing growth characteristic. There is no public data on the DAU to MAU ratio, making it impossible to make a definitive judgment on user stickiness.
Yuanbao's role in Tencent AI's path is "C-end validation of independent products." It has proven two things: first, Tencent can leverage WeChat's social chain to bring AI products to a user base of tens of millions; second, users attracted by red envelopes cannot be retained. Liu Chiping stated at the earnings meeting that Yuanbao's Spring Festival promotional effects exceeded expectations and that the next step would focus on optimizing core capabilities such as voice interaction; this statement itself also shows that the team understands that retention is the core proposition of the next stage.
The pulsing growth experience of Yuanbao, in turn, explains why WeChat AI chose to integrate natively within the super application instead of continuing to push an independent app. Independent apps require users to actively open them, with retention relying on push notifications and activities; native integration relies on scenario binding, such that when users need to order, pay, or check deliveries, WeChat AI is right in the conversational flow. This represents two completely different retention logic.
Every Mini-Program Can Be "Lobsterized," But Service Providers Fear Being Short-Circuited
The product direction of WeChat AI has a clear outline as articulated by Ma Huateng in March 2026.
In the 2025 earnings report communication meeting, Ma Huateng first discussed the concept of "raising shrimp." The "lobster" applications he referred to are AI agents with a "human-like" feeling capable of executing tasks autonomously rather than just answering questions. Ma Huateng indicated that such applications inspired the planning of WeChat AI: in the future, every mini-program could achieve intelligent and "lobster-like" transformation.
The core of this metaphor is to promote AI from a conversation tool to a task executor. If WeChat AI is just a chatbot, it does not need to read the source code or operate pages. The existence of the automatic mode indicates that its positioning is to complete cross-mini-program tasks for users: ordering a cup of coffee, paying a water and electricity bill, making a hospital appointment, or starting a mini-game. Users do not need to know which mini-programs provide these services; they just need to say a sentence to WeChat AI.
However, Ma Huateng proactively mentioned the contradiction of ecological interests in the same meeting. He pointed out that ecosystem service providers are worried about being "short-circuited" and "channeled" by AI intelligences. If a user tells WeChat AI, "Help me order a latte," and AI directly invokes a particular coffee mini-program's atomic service to complete the transaction without the user entering the merchant's page at all, then the merchant's advertising position, brand exposure, and user retention would all be eliminated. Service providers would not accept this outcome.
This is the core contradiction in WeChat AI product design. The more efficient centralized scheduling is, the weaker the decentralized flow authority of merchants becomes. The two access modes do not resolve this contradiction themselves; they simply represent an entrance design. The real balancing mechanism, such as traffic distribution rules, the relationship between atomic services and merchant pages, and data visibility for service providers, have not been publicly disclosed at all. Ma Huateng's original words were, "We must balance centralized scheduling and decentralized flow protection," but how to balance them has not yet been answered during the internal testing phase.
Three Lines Have Been Established, but the Third Step Has Only Just Begun
With Mixian, Yuanbao, and WeChat AI advancing simultaneously, Tencent AI's gradual path is logically coherent.
Not pursuing the fastest models in the underlying layer, but rather building the most stable Agent foundation. Mixian's ranking as number one in application capability in SuperCLUE supports the operational accuracy demand of WeChat AI. The middle layer runs a separate app to navigate social chains for new user acquisition and basic experience; Yuanbao's MAU exceeding one hundred million during the Spring Festival validated the leverage effect of WeChat's traffic pool for AI products. The upper level integrates natively within the super application, using scenarios to reduce retention pressure, as WeChat AI's internal testing directly faces 400,000 developers and a mini-program ecosystem with daily active users exceeding 100 million.
However, whether C-end perception has been turned around can currently only be judged as "partially completed." Yuanbao's monthly active users in the hundreds of millions primarily came from the pulsing red envelope influx, with normal DAU around 9 million, which is still a significant gap compared to Doubao's 14 million. WeChat AI has just entered internal testing, and ordinary users are still unable to perceive it. Tencent AI's share of public consciousness shows a noticeable discrepancy from its technical level.
Whether WeChat AI can bridge this gap depends on three variables. First, whether the source code trust issues in the automatic mode can be resolved on the developer side, which determines the supply-side access scale. Second, whether the centralized and decentralized traffic distribution rules can be accepted by service providers, which dictates whether ecological interests can be balanced. Third, whether the accuracy of AI operations and liability attribution can reassure users enough to place orders, which determines the depth of C-end use.
Establishing three lines is a prerequisite, but whether they can form a chain of "Mixian ensuring reliability, Yuanbao validating user habits, WeChat AI delivering the final experience" will require at least two quarters of public data for verification. Ma Huateng mentioned in the earnings meeting that "AI is a marathon, not a sprint," and the internal testing of WeChat AI is just a marker in the mid-point of this marathon, with a long way to go before reaching the finish line.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
