From Kling to HappyHorse: Zhang Di's 5-Month Journey

The Kling Legacy

Zhang Di is widely known in the AI video community as the "father of Kling AI." As Vice President at Kuaishou, he served as the technical leader for Kling, one of the first Chinese AI video models to gain international recognition and compete on global leaderboards.

Kling AI established itself as a serious competitor in the AI video generation space, particularly for its image-to-video capabilities. Under Zhang Di's technical leadership, the model achieved strong rankings on the Artificial Analysis Video Arena and gained a significant user base.

The Move to Alibaba

In November 2025, Zhang Di departed Kuaishou. According to Atlas Cloud's reporting, he joined Alibaba's "Future Life Lab," initially part of Taobao's research division, reporting to Chief Scientist Zheng Bo.

The timing was significant. Alibaba was in the process of a major AI reorganization. In March 2026, CEO Eddie Wu established the "Alibaba Token Hub" (ATH), consolidating five previously separate AI units — including Tongyi Laboratory, Qwen, and the AI Innovation Unit — under his direct oversight.

Zhang Di's team was moved from the Future Life Lab into the AI Innovation Unit within ATH. This gave the team direct access to Alibaba's compute resources and a clear mandate to build competitive AI products.

Five Months to #1

From Zhang Di's arrival at Alibaba in late 2025 to HappyHorse-1.0's appearance on the Artificial Analysis leaderboard on April 8, 2026, approximately five months elapsed. In that time, his team built a 15-billion-parameter unified Transformer model that:

• Topped the Artificial Analysis Video Arena in text-to-video and image-to-video
• Achieved the highest Elo score ever recorded on the platform (1413 in image-to-video)
• Implemented simultaneous audio-video generation in a single inference pass
• Supported native lip-sync for 7 languages
• Beat both his former creation (Kling) and ByteDance's Seedance 2.0

The "Quiet Launch" Strategy

HappyHorse's debut was notably unconventional. When it first appeared on the Artificial Analysis leaderboard on April 8, there were no press releases, no blog posts, and the company name was obscured. The model page simply read "coming soon."

Atlas Cloud notes that this "quiet launch followed by a spectacular announcement" is becoming a trend among Chinese AI companies. Xiaomi used a codename "Hunter Alpha" and Zhipu used "Pony Alpha" for similar stealth launches of their models.

The strategy generated significant speculation and discussion in the AI community before Alibaba officially confirmed the model's origins on April 10 — two days after the leaderboard appearance.

The Dual-Engine Structure

According to Atlas Cloud's analysis, HappyHorse demonstrates that ATH possesses what they describe as a "dual-engine structure":

• Tongyi Lab: Focused on foundational research (responsible for Qwen language models)
• AI Innovation Unit: Building applications from real business challenges (responsible for HappyHorse)

This structure means Alibaba now has two separate teams capable of producing top-tier multimodal models, each with different strengths and approaches.

What Comes Next

The HappyHorse-1.0 API is scheduled for public release on April 30, 2026. The model is fully open-source with no commercial restrictions. Fine-tuning scripts have not yet been released but have been hinted at for the future.

For Zhang Di personally, the success of HappyHorse validates his move from Kuaishou to Alibaba. Having built two separate #1-ranked AI video models at two different companies, he has established himself as one of the most consequential figures in the AI video generation field.