On June 8, 2026, the WeChat Developer Platform announced that WeChat AI has entered the beta testing phase. This AI assistant, integrated within the WeChat ecosystem, enables users to directly invoke, access, and interact with mini-programs through natural language conversations. The open platform offers two integration modes: the automatic mode allows the platform to read mini-program source code, enabling AI to directly operate pages without additional development; the development mode lets developers build custom skills independently, which are then made available for AI invocation after review. The terms of service also note that “WeChat AI” may be a temporary name, as the final name has not yet been determined, and integration is optional, with no impact on the normal operation of existing mini-programs.
This is the first time WeChat has opened its mini-program ecosystem to AI at the conversation entry level. The context is: Tencent’s proprietary HunYuan large model has now ranked among China’s top-tier models in public benchmark tests, and the Yuanbao app has surpassed 100 million monthly active users following the 2026 Spring Festival red packet surge. WeChat’s AI beta represents Tencent’s latest step in transitioning its AI capabilities from technical groundwork and standalone product validation to delivery through a super-app. The automatic mode requires developers to submit their source code—this low-barrier pathway could attract how many developers, and what ecosystem conflicts might it trigger? These are the key questions the beta phase aims to answer.
The conversation layer entry point of the mini-program ecosystem
WeChat AI offers two integration modes, each targeting entirely different developer communities.
The design logic of automatic mode is straightforward: upon approval, the platform is authorized to access the mini-program's source code and automatically analyzes the page structure, enabling AI to interact with the page without any additional development. A small game team of just two or three people doesn’t need to hire an AI engineer or understand Agent protocols—by simply enabling authorization, their ordering mini-program or utility app can be invoked by WeChat AI.
According to data disclosed by WeChat Open Class in January 2026, the WeChat mini-game ecosystem has gathered over 400,000 developers, 80% of whom are teams of fewer than 30 people. In 2025, daily active users exceeded 100 million, and monthly active users surpassed 500 million. This scale on the supply side is a unique moat exclusive to WeChat AI. While ByteDance’s Doubao or Alibaba’s Tongyi Qianwen can develop standalone apps or open APIs, they lack a mini-program ecosystem with over 100 million daily active users for direct integration. WeChat AI’s automatic mode essentially trades technological convenience for large-scale integration, enabling the vast majority of these 400,000 developers to onboard at zero cost.
The development mode provides customization opportunities for service providers with complex business logic. Developers can independently build skills based on their own business characteristics, which, after evaluation and review by the platform, can be invoked by WeChat AI. Both modes can be enabled simultaneously and are not mutually exclusive.
The phrasing “name not yet determined” and “optional behavior” indicates that the WeChat team still maintains reservations about the product’s positioning. The primary goal during the beta phase is to validate the technical workflow and observe developer reactions. However, the automatic mode has already touched on a sensitive issue: source code authorization. Some developers have expressed concerns on the WeChat Open Community, with core issues centered on several points—how platform access to source code ensures code asset security, whether AI-operated page interactions could disrupt existing tracking and ad display logic, and how liability should be assigned if AI errors result in user losses. No public guidelines currently address these concerns.
After ranking second domestically in foundational capabilities, Hunyuan chose to go deeper.
WeChat AI requires more than just a conversational model—it needs an agent foundation capable of understanding page structures and precisely executing operational commands. This foundation is the Tencent HunYuan large model.
In March 2025, the Chinese large model evaluation benchmark SuperCLUE released a report showing that Tencent HunYuan Flagship ranked second nationally among foundational models, behind ByteDance’s Doubao, but ranked first nationally in application capabilities, leading in subcategories such as text understanding and generation, instruction following, and Agent capabilities. ScienceNet, summarizing the report, noted that HunYuan outperformed its foundational model ranking in the “practical application” dimension. Around the same time, HunYuan Turbo S made its debut in the global Top 15 of the international evaluation platform Chatbot Arena.
Hunyuan follows a quarterly release cadence. In April 2025, hunyuan-turbo was updated, and in July, the flagship TurboS was launched with enhanced reasoning capabilities. In April 2026, the Hy3 preview version was released, with the official claim of a 40% improvement in inference efficiency. According to Tencent Cloud product documentation, older versions such as HY 2.0 are scheduled to be discontinued effective June 26, 2026.
This pace is much slower than ByteDance and Alibaba. ByteDance’s Doubao and Alibaba’s Qwen maintained a model release frequency close to “weekly updates” over the past year, while Hunyuan has consistently delivered major version updates on a quarterly basis. Tencent’s management has previously publicly stated a preference for “slow and meticulous work.” From a technical standpoint, the explanation is that the Agent era demands far greater stability and lower latency than the conversational era, and frequently switching underlying models makes it difficult for developers to perform engineering adaptations. For WeChat AI, use cases include operations involving funds and sensitive information—such as placing orders, paying bills, and making appointments—where model output consistency is far more important than creativity.
Regarding resource allocation, Tencent President Liu Chiping disclosed at the 2025 annual report briefing that the company invested RMB 18 billion in AI product development in 2025, and this investment will at least double in 2026. According to reports from The Paper, Liu also stated that the next key strategic focus is to build a dedicated AI agent within WeChat, integrating the full ecosystem of mini-programs, social features, and payments. The doubling of investment without accelerating release timelines indicates that the additional funding is primarily directed toward rebuilding infrastructure and improving data quality, rather than rushing to market.
Hunyuan’s leadership in application capabilities aligns with the practical needs of WeChat AI. A model that ranks higher based on foundational metrics but has weaker agent capabilities may actually be less effective in WeChat AI scenarios than Hunyuan. Tencent has chosen a path that avoids parameter competition and focuses on practical performance—a strategy that has demonstrated internal consistency since the early testing phase of WeChat AI.
Spring Festival daily active users exceed 50 million—so what?
Before the WeChat AI beta, Tencent AI's C-end validation tasks were handled by the Yuanbao app.
The growth curve of Yuanbao exhibits distinct pulse characteristics. According to data from QuestMobile, as reported by China National Radio, Yuanbao's monthly active users ranked 12th in the industry in January 2025, rising to 3rd place by December 2025, behind DouBao (MAU: 226 million) and DeepSeek (MAU: 135 million), with a full-year compound growth rate of 27.8%.
During the 2026 Spring Festival, Yuanbao experienced explosive growth. Official data from Tencent revealed that Yuanbao’s daily active users (DAU) peaked at over 50 million, reaching 40.54 million on New Year’s Eve, with monthly active users (MAU) hitting 114 million. The Shanghai Securities News reported that this growth was primarily driven by social referral through red packet activities.
However, data rapidly declined after the Spring Festival. According to QuestMobile, in April 2026, Yuanbao’s regular DAU was approximately 9 million, while DouBao’s DAU was around 140 million and Qianwen’s DAU was about 30 million. The peak-to-trough difference approached fivefold, indicating a clear pulsating growth pattern. Public data on the DAU-to-MAU ratio is unavailable, making it impossible to determine user retention with certainty.
Yuanbao's role in Tencent's AI roadmap is "C-end validation for an independent product." It demonstrates two things: first, that Tencent has the capability to leverage WeChat's social network to bring AI products to hundreds of millions of users; second, that users acquired through red packets cannot be retained. Liu Chiping stated on the earnings call that Yuanbao's Spring Festival promotion exceeded expectations, and the next focus will be on optimizing core capabilities such as voice dialogue. This statement itself indicates that the team recognizes retention as the central challenge for the next phase.
Yuanbao's pulse-based experience growth illustrates why WeChat AI chose to natively integrate within the super app rather than continue promoting standalone apps. Standalone apps require users to actively open them, relying on push notifications and promotions for retention; native integration, by contrast, binds users through context—when users need to order food, pay bills, or check deliveries, WeChat AI is already present within the conversation flow. These are two entirely different retention strategies.
Every mini-program can be "lobster-ized," but service providers fear being short-circuited.
The product direction of WeChat AI was clearly outlined in Ma Huateng's public statement in March 2026.
Content from the 2025 annual report briefing: Ma Huateng first discussed the concept of “shrimp farming.” The “lobster-type” applications he referred to are AI agents with a “human-like presence” that can autonomously execute tasks rather than merely answer questions. Ma stated that these applications have inspired the development of WeChat’s AI: in the future, each mini-program could be intelligently transformed and “lobster-ized.”
The core of this analogy is to elevate AI from a conversational tool to a task executor. If the WeChat AI were merely a chatbot, it wouldn’t need to read source code or interact with interfaces. The existence of the automatic mode reveals its purpose: to perform cross-miniprogram tasks on behalf of users—ordering a coffee, paying a utility bill, booking a hospital appointment, or launching a mobile game. Users don’t need to know which specific miniprogram provides each service; they simply need to speak one sentence to WeChat AI.
But Ma Huateng voluntarily raised the issue of conflicting interests within the ecosystem during the same conference. He noted that ecosystem service providers fear being “bypassed” or “channelized” by AI agents. If a user tells WeChat’s AI, “Help me order a latte,” and the AI directly invokes an atomic service from a coffee mini-program to complete the transaction without the user ever entering the merchant’s page, the merchant’s advertising space, brand exposure, and user retention all vanish. Service providers will not accept this outcome.
This is the core contradiction in WeChat's AI product design: the more efficiently centralized orchestration operates, the weaker merchants' decentralized control over traffic becomes. Neither of the two integration models resolves this contradiction—they are merely entry-point designs. The true balancing mechanisms, such as traffic allocation rules, the relationship between atomic services and merchant pages, and data visibility in service provider backends, have not been publicly disclosed at all. Ma Huateng's exact words were, “Centralized orchestration and decentralized traffic protection must be balanced,” but how exactly to achieve this balance has not been answered during the beta phase.
The three lines are in place, but step three has just begun.
With Hunyuan, Yuanbao, and WeChat AI advancing in parallel, Tencent’s incremental AI strategy is logically consistent.
At the foundational level, we don’t aim to build the fastest model—we build the most stable agent base. Hunyuan’s top ranking in application capabilities on SuperCLUE in China supports WeChat AI’s demand for precise operations. At the middle layer, a standalone app successfully validated social chain-driven user acquisition and core user experience, with Yuanbao’s MAU exceeding 100 million during the Spring Festival, proving WeChat’s traffic pool leverages AI products effectively. At the top layer, we integrate natively within the super-app, using contextual scenarios to reduce retention pressure; WeChat AI’s internal testing directly engages 400,000 developers and a mini-program ecosystem with over 100 million daily active users.
However, whether end-user perception has reversed can only be judged as “partially achieved” at this stage. Yuanbao’s hundreds of millions of monthly active users primarily stem from red packet spikes, with a normal DAU of approximately 9 million—still significantly behind DouBao’s 140 million. WeChat AI has just entered internal testing, and general users have yet to perceive it. There remains a clear gap between Tencent AI’s share of public awareness and its technological capabilities.
Whether WeChat AI can bridge this gap depends on three variables. First, whether the trust issue with the source code of the automated mode can be resolved by developers, which determines the scale of supply-side integration. Second, whether the centralized and decentralized traffic allocation rules can be accepted by service providers, which determines if ecosystem interests can be balanced. Third, whether the accuracy and accountability of AI operations can reassure users to place orders, which determines the depth of end-user adoption.
The alignment of the three lines is a prerequisite, but whether they can form a chain—“Hunyuan ensuring reliability, Yuanbao validating user habits, and WeChat AI delivering the final experience”—requires at least two quarters of public data to verify. Ma Huateng said on the earnings call, “AI is a marathon, not a sprint”; the WeChat AI beta is merely a milestone midway through this marathon, with a long way still to go before the finish line.