Alibaba's ATH has introduced a new member to its "Happy Family" of AI products. Shortly after HappyHorse gained global attention with its video generation capabilities, re-establishing Alibaba's prominence in the multimodal AI arena, the company swiftly revealed another strategic product. On April 16, 2025, Alibaba launched its "open-world model product," HappyOyster. While the name follows the branding system of HappyHorse, its focus is distinctly different. If HappyHorse addresses "how to generate better videos," HappyOyster attempts to answer "how to generate a persistently existing, real-time interactive world." Coincidentally, on the very same day, Tencent officially released and open-sourced its mixed-reality 3D "world model," HY-World 2.0, emphasizing 3D scene generation compatible with game engines and the construction of interactive worlds. The near-simultaneous bets by these two Chinese internet giants on the "world model" narrative signal that AI competition is entering a new phase: evolving from answering questions and generating text to constructing spaces, simulating reality, and creating interactive digital worlds.

The AI video generation industry has, over the past year, established a familiar workflow: input a prompt, wait for several seconds or minutes, and receive a video clip lasting a few seconds to a minute. Users act like directors or clients, submitting requests and awaiting the machine's output. This model addresses content production efficiency but remains fundamentally a one-way generation process. HappyOyster aims to disrupt this logic. Alibaba defines it as a "world model" because it allows users to continuously issue commands during the generation process, with the visuals responding in real-time and the world evolving persistently. Users can move perspectives, alter environments, direct characters, and rewrite narratives within a generated space. The model does not deliver a final product in one go but operates continuously. Consequently, users transition from being mere prompt engineers to becoming participants, controllers, and even co-creators of this world. Strictly speaking, however, HappyOyster is still built on a generative model framework and is not yet the fully realized, self-consistent, universal world model with advanced physical laws and causal logic that some might envision. It represents more of a transitional form: a step beyond video generation toward real-time simulation and continuous interaction. Limitations remain, such as finite visual detail, latency in movement, and room for improvement in character consistency and command comprehension, indicating its early-stage development. Yet, the crucial aspect is not its current perfection but the clear direction it sets.

Previously, industry discussions on AI were predominantly focused on large language models—comparing parameters, reasoning capabilities, coding accuracy, and potential to replace office software. As we move into 2026, the impact of such narratives is beginning to diminish. While text capabilities continue to improve, user novelty is waning, and capital markets increasingly recognize that standalone chatbots are unlikely to justify high valuations. The true potential for opening new markets lies in multimodality, video, interactivity, and digital worlds. Alibaba has evidently anticipated this shift. The sequential launches of HappyHorse and HappyOyster, while appearing as two separate products, are actually two strategic moves along the same path. The first step was to demonstrate global competitiveness in video generation, recapturing market attention. The second step elevates the competition from video generation to world generation, upgrading Alibaba's AI narrative from "capable of building models" to "capable of defining the next generation of interaction paradigms." This reflects a broader acceleration in Alibaba's AI initiatives. Recently, the company reorganized its internal AI structure, establishing the ATH (Alibaba Token Hub) business group, which consolidates the Tongyi Lab, MaaS business line, Qianwen division, Wukong division, and AI Innovation division under a unified system. While external attention initially focused on the organizational changes, the true significance is the consolidation of resources previously scattered across cloud, model, application, and innovation teams, with the singular goal of accelerating AI product development and commercialization. HappyOyster was born from this context, emerging from the Innovation division under ATH, the same team responsible for HappyHorse. This is noteworthy. In recent years, Alibaba's most recognized AI assets were Tongyi Qianwen and Alibaba Cloud, with innovation projects often perceived as experimental sidelines. Now, however, Alibaba's most aggressive and talked-about products are originating from its Innovation division. This indicates the emergence of a new internal mechanism: mature businesses focus on steady growth, while innovation teams are tasked with betting on future gateways. This division of labor resembles Alibaba's historical approach in e-commerce. After Taobao matured, Tmall addressed brand elevation, Xianyu handled the long tail, and Hema explored offline retail. Today, Alibaba is applying this logic to the AI era: Tongyi manages foundational models, Alibaba Cloud handles commercial deployment, Qianwen serves as the public gateway, while products like HappyHorse and HappyOyster are tasked with exploring future interaction paradigms.

From an industry perspective, the emergence of HappyOyster carries another significant implication: it marks the point where major Chinese tech firms begin direct competition in the interactive generation domain. Tencent's newly launched HY-World 2.0 emphasizes industrial application. It can generate assets in various formats like Mesh, point clouds, and 3D Gaussian Splatting (3DGS), and integrates with existing game engine workflows, enabling developers to rapidly construct maps, levels, and prototype scenes. Tencent's strategy is clear: first deploy the model within its core strength—the gaming business—to address challenges like lengthy open-world development cycles, high art costs, and insufficient content production capacity. This aligns with Tencent's pragmatic calculus. As the world's highest-grossing game company, Tencent possesses vast 3D data, mature engine expertise, and continuous content demands. A mature model would primarily benefit its game R&D ecosystem; tasks that previously took months, like creating map drafts, might eventually be generated from a single prompt, and prototype development requiring large art teams could be handled by smaller teams augmented by AI. Alibaba's approach is slightly different. HappyOyster prioritizes real-time interaction, world evolution, and an open creative experience. Rather than initially targeting industrial toolchains like Tencent, it leans closer to being a next-generation user interaction gateway. Tencent aims to solve production problems first, while Alibaba is experimenting with the future form of consumer-grade experiences. Neither path is superior; they simply start from different points. Tencent enters from the gaming industry, Alibaba from content interaction, but both may be converging on the same destination: AI as a real-time generation engine for digital worlds. This also signifies a shift in the competitive logic among Chinese tech giants in the AI arena. The previous phase centered on which general-purpose model had more parameters or higher benchmark scores. The current phase is about which company can identify the implementation direction that synergizes most deeply with its core business. Tencent is betting on 3D worlds because gaming is its cash cow; Alibaba is betting on multimodal interaction because its e-commerce, content, and cloud computing businesses all require richer digital space gateways.

Viewed from this angle, even in its nascent state, HappyOyster warrants serious attention. It reveals not product maturity, but corporate direction. The fact that a major player is willing to invest resources in such a highly uncertain frontier product indicates an internal consensus: the next stage of AI involves not just generating content, but generating environments; not just answering questions, but facilitating actions. For Alibaba, this step is particularly critical. In recent years, external discussions about Alibaba often centered on e-commerce competition, slowing cloud growth, and frequent organizational adjustments, while its AI narrative was overshadowed by ByteDance, Tencent, Baidu, and even startups. However, in recent months, Alibaba has clearly been striving to regain initiative. CEO Yongming Wu explicitly set a target for cloud and AI annual revenue to exceed $100 billion within five years, organizationally elevated AI to a core group position, and intensified the release of new models and product forms. HappyHorse served as a demonstration of technical capability; HappyOyster demonstrates strategic ambition. This is because video generation remains a competition within an existing market, whereas interactive generation represents a potential增量 market. It connects not only to content creation but potentially to gaming, education, cultural tourism, virtual socializing, robot training, and even the interaction systems of future spatial computing devices. The company that positions itself effectively in this phase will have the opportunity to help define the next-generation computing platform. Alibaba evidently aspires to be more than just a model supplier within someone else's ecosystem. While today's HappyOyster might still be a somewhat rudimentary "Happy Oyster," and Tencent's HY-World 2.0 remains in its early polishing stages, the fact that both giants played their cards on the same day underscores a reality: interactive generation is no longer just a laboratory concept but is becoming a key battleground for Chinese tech companies. The real competition has just begun.