Tencent's AI image model surges 8% on high volume, mirroring the success of Meta Muse.

Tencent Holdings' Hong Kong-listed shares surged nearly 8% on Tuesday, with trading volume soaring to HK$17.7 billion. The catalyst came from two interconnected news items: Meta's AI agent app Muse rapidly gained popularity in the US, topping app store download charts, while Tencent released its latest image generation model, Hy Image 3.5 Preview, on the same day. This created a clear market correlation – Tencent, with its super social ecosystem and access to high-frequency everyday scenarios, is seen as the most suitable platform for the "Chinese version of Muse."

Meta Muse entered the top three of the US App Store's free charts on its second day of release and has consistently remained high on productivity charts. Muse's core difference lies in going beyond the scope of a chatbot: by reading social relationships and integrating various APIs such as email, calendar, shopping, and food ordering, it asynchronously and autonomously performs complex cross-application tasks for users in the background.

This blockbuster effect prompted the capital market to turn its attention to domestic counterparts. With its massive mini-program service ecosystem and closed-loop payment system, WeChat is considered to have an efficiency of implementation and commercial monetization potential that is no less than that of Meta once the AI Agent is deeply integrated. As a result, Tencent became the most sought-after beneficiary company that day.

A research report from CICC points out that Meta Muse's breakthrough has shifted from "whether the model is smart enough" to "whether users can confidently deliver products and whether the product can connect to real-world transactions." The competitive focus of 2C agents has officially shifted from the technology layer to ecosystem distribution and trust architecture. Gary Tan, portfolio manager at Allspring Global Investments, stated, "Some investors are drawing parallels between Tencent and Meta, especially considering WeChat's unique social ecosystem and its potential to support large-scale personal AI assistants."

Tencent released its image model on the same day, accelerating its pace of development.

Tencent's release of Hy Image 3.5 Preview is its latest move in the direct competition of the AI image generation field. In its statement, Tencent said the model offers improved performance compared to its predecessor and has been integrated into its Yuanbao product line, as well as video editing and design tools.

Tencent used hundreds of its internal designers as a testing group and stated that the model performed comparably to ByteDance's Seedream 5.0 Pro, while slightly outperforming Alphabet's Google's Nano Banana Pro and Alibaba's Qwen-Image-3.0 Pro. Tencent has thousands of game designers and artists, who are expected to be the primary beneficiaries of the tool; however, Tencent did not provide quantifiable details regarding its quality claims.

The timing of this announcement, coinciding with the launch of Alibaba's AI conference, is quite striking. Since hiring former OpenAI researcher Yao Shunyu as Chief AI Scientist, Tencent's AI strategy has shifted: Yao publicly criticized the practice of training models solely for leaderboard scores, emphasizing product integration and solving real-world problems. Another former OpenAI researcher and computer vision expert, Tian Yonglong, also joined Tencent's Hunyuan team in July, responsible for visual language model development.

The logic behind Muse's blockbuster success: a paradigm shift from chat to delivery.

To understand the core of Tencent's recent surge, it is necessary to clarify what new narrative Meta Muse has opened up.

A CICC research report points out four key differences between Muse and previous similar agent products: First, it executes asynchronously in the background, retains long-term memory and user preferences across conversations, and tasks can continue even after the user closes the application; second, it connects to real accounts, accessing scenarios such as email, calendar, payment, health, e-commerce, and smart home through three methods: built-in connectors, public APIs, and browser operations; third, it features isolated execution and approval gating, with each user having an independent Muse Secure VM runtime environment, and Sentinel acting as an agent to safeguard all external operations, making security architecture a core part of the product design; and fourth, the Ideas and Feed functions reveal the rudiments of a recommendation-based agent, with Muse shifting from waiting for user questions to proactively identifying needs and defining problems in advance.

CICC's research report argues that the ultimate competition in the 2C agent market will resemble "recommendation" rather than "search"—the agent will proactively identify needs based on sufficient context, define and solve problems for users, rather than passively waiting for instructions. Meta's advantage lies in its distribution and trust infrastructure: 3.6 billion daily active users provide a large reach, WhatsApp conversation threads lower the barrier to entry, and the long-term accumulation of content behavior on Instagram and Facebook forms a richer profile of personal needs, which is difficult for products like ChatGPT, which rely solely on chat history, to replicate.

In terms of business model, Muse adopts a free tier and two monthly subscription tiers of $20 and $100, with the highest tier pricing approaching the level of some B2B software. A research report from CICC believes that as Muse undertakes consumption tasks such as shopping and reservations, its monetization path is expected to form a hybrid model of "subscription as a base + transaction commission".

Why Tencent is the optimal solution for the "Chinese version of Muse"

The capital market's comparison of Tencent to Meta Muse is not a simple analogy; it is supported by a clear ecosystem logic.

A CICC research report, analyzing the differences between the 2C agent ecosystems in China and the US, points out that WeChat has integrated accounts, payments, and lifestyle services within its platform and possesses a unique mini-program ecosystem, which allows its own agents to directly access capabilities and complete transactions within authorized scope. In contrast, Meta's overseas service entry points are relatively fragmented, with web pages remaining a crucial carrier. Muse relies on browser operations to cover long-tail services, resulting in relatively high connection and maintenance costs, and website redesigns may also impact execution efficiency.

In other words, if Tencent deeply integrates its AI Agent into the WeChat ecosystem, its inherent closed-loop mini-program ecosystem and payment system will give it lower execution friction than Meta. Tencent's vision—to build an AI capable of performing various tasks for over 1 billion users in the WeChat ecosystem—aligns perfectly with Muse's product direction. This is the core logic behind the market's overpricing of Tencent that day.

Timing determines rhythm: two paths and four stages

CICC's research report introduces a product innovation cycle framework, dividing the development of 2C agents into four stages: exploration stage (both technical effectiveness and cost have not reached the threshold), transitional innovation stage (technical effectiveness has reached the threshold but cost is still high), all-round innovation stage (both effectiveness and cost have broken through the threshold), and the strong get stronger stage (the technology curve tends to flatten).

The report uses the historical paths of BlackBerry and iPhone as a reference: the transitional innovation period corresponds to BlackBerry's strategy of focusing on enterprise email and cutting non-core functions, with BlackBerry's revenue increasing from $600 million in 2004 to a peak of $20 billion in 2011; the all-around innovation period corresponds to the all-in-one integration of iPhone, which has an ecosystem that crushes the transitional products.

CICC's research report believes that the continuous improvement of AI model capabilities and the decline in token computing power costs are currently the most certain factors. More difficult to grasp are users' tolerance for errors in agent-based task completion, their willingness to migrate from existing habits, and users' cost thresholds. Based on this, there are two possible development paths for 2C agents: one is a transitional innovation phase, starting with basic office needs and gradually extending to comprehensive scenarios; the other is a rapid decline in both technology and cost curves, directly skipping the transition period and entering a comprehensive innovation phase—at which point giants like Meta and Tencent, with their distribution advantages and deep ecosystems, will benefit first. CICC's report emphasizes that the shorter the interval from the transitional phase to the comprehensive phase, the greater the advantage for internet giants; conversely, startups have more time to build barriers to entry.

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.