Alibaba is simultaneously operating two competing video generation models on the same platform, a rare strategy that signals internal tension between competing teams and hints at a larger consolidation to come. On August 24, Alibaba completed a 10.2 billion U.S. dollar financing round and fully launched Wan3.0, its latest video generation model, just weeks after the company had already claimed HappyHorse, another advanced video tool that topped global AI benchmarks in April.

What Makes Alibaba's Two-Model Strategy Unusual?

Most major AI companies maintain a single public-facing video model. ByteDance exports its multimodal capabilities as Seedance, while Kuaishou offers Kling as its unified video generation tool. Alibaba, by contrast, is allowing Wan3.0 and HappyHorse to compete directly for the same customers on its Bailian platform, with nearly identical feature sets but different pricing structures.

The two models share remarkable functional overlap. Both support 30-second video generation, full-modality reference inputs, native audio integration, and character consistency features. However, Wan3.0 costs 0.6 yuan per second for 720P video, while HappyHorse 1.1 charges 0.9 yuan per second, making Wan3.0 the cheaper option despite matching capabilities. According to Alibaba Cloud agents, both models primarily target e-commerce customers, though neither has significantly dented Seedance's dominance in short-form drama and comic content.

How Did Alibaba End Up With Two Competing Video Models?

The answer lies in Alibaba's internal reorganization and the different origins of each team. Wan3.0 emerged from Tongyi Wanxiang, a multimodal team founded in 2015 by Liu Yu that had been iterating on video generation since earlier versions like Wan2.1 through Wan2.7. HappyHorse, by contrast, originated in Alibaba's Taobao and Tmall Technology Department before being reassigned multiple times through organizational restructuring.

The timing of Wan3.0's launch was strategically delicate. In June, Alibaba established the Token Foundry Business Unit by merging the Tongyi Large Model Business Unit with the Future Life Laboratory. One month later, in July, the Tongyi Wanxiang team was officially merged into the Future Life Laboratory, which is led by Alibaba Vice President Zhang Di and serves as the research hub for HappyHorse. This consolidation placed three multimodal models under Zhang Di's leadership: Wanxiang (Wan3.0), HappyHorse, and HappyOyster, a world model.

Despite the organizational merger, the teams have not fully integrated their operations. According to an Alibaba employee cited in the source material, the Wan3.0 team recently relocated from Alibaba's Yungu office area to Xixi Park, but the two businesses continue advancing separately. The teams have begun collaborating on training HappyHorse 1.2, signaling potential future convergence.

Why Alibaba Might Consolidate These Models

Industry observers expect eventual consolidation. According to reporting from Yiyu Observation cited in the source material, Alibaba has put the merger of Wan and HappyHorse on the table, with HappyHorse likely to become the primary model going forward. This mirrors how other Chinese tech giants have handled competing AI products. Doubao, for example, integrated Feishu, TRAE, and Coze into a unified offering.

However, HappyHorse faces real limitations in the market despite its technical specifications. The model boasts 15 billion parameters, nearly three times that of Seedance, and supports 15-second multi-shot narrative generation with multi-aspect ratio adaptation and 1080P super-resolution output. Yet short-drama practitioners have reported significant shortcomings in character facial texture, with severe skin smoothing and unrefined skin details that make the output unsuitable for high-quality short-form content. One short-drama creator noted that HappyHorse performs better in e-commerce scenarios, such as tea-picking advertisements, but struggles with complex shots.

Steps to Understanding Alibaba's AI Video Strategy

• Model Specifications: Wan3.0 generates 30-second videos and now supports document inputs like Word, Excel, PowerPoint, PDF, and Markdown files, positioning it as an enterprise-level content generation tool rather than just a consumer video maker.
• Pricing Dynamics: Wan3.0 undercuts HappyHorse by 33 percent on per-second costs, giving Alibaba Cloud agents incentive to recommend the newer model to cost-conscious e-commerce customers.
• Market Positioning: Both models primarily compete in e-commerce scenarios and have not significantly challenged Seedance's dominance in short-drama and comic content, suggesting Alibaba is targeting a different customer segment.
• Organizational Integration: The Wan and HappyHorse teams operate under the same leadership but maintain separate operations, indicating a transition period before potential full consolidation.

The broader context matters here. HappyHorse was launched in April 2026 when Seedance 2.0 was at peak popularity, and the model was initially regarded as Alibaba's answer to ByteDance's dominance. Zhang Di, who leads the Future Life Laboratory and oversees HappyHorse, is known as the former "Father of Kling," Kuaishou's video generation model, adding intrigue to his role in developing Alibaba's competing offerings. However, HappyHorse's trajectory has been described as a "high start and low go" curve, generating initial excitement when it anonymously topped AI evaluation benchmarks but failing to exceed expectations during gray testing.

Alibaba's decision to launch Wan3.0 while HappyHorse remains in development suggests the company is hedging its bets in a rapidly evolving market. The strategy allows Alibaba to serve different customer segments and price points while maintaining optionality about which model to promote long-term. Whether this dual approach strengthens Alibaba's position or signals internal uncertainty about which technology will ultimately prevail remains an open question as the video generation market continues to consolidate around a handful of dominant players.