Crushing Seedance 2.0, Alibaba announces: "Happy Horse" video model is undergoing internal testing.

Alibaba has officially "claimed" the mysterious video generation model HappyHorse, which has sparked widespread speculation in the AI community. This marks Alibaba's most significant achievement in the video AI field to date and is the latest sign of its accelerated AI strategy.

On April 10, Alibaba ATH confirmed that HappyHorse is a model developed by its Innovation Division and is currently in internal testing. The API will be open for access soon, reportedly scheduled for April 30.

The model previously topped the text-to-video leaderboard on the globally renowned AI evaluation platform Artificial Analysis, anonymously pushing ByteDance’s Seedance 2.0 to second place and igniting massive speculation among the AI community and investor circles.

HappyHorse's sudden emergence directly boosted Alibaba's stock price—during the news fermentation, Alibaba's shares rose more than 4% intraday. Jefferies analyst Thomas Chong wrote in a research report that HappyHorse is "a success" for Alibaba.

Topping the Leaderboard, Crushing the Competition

HappyHorse-1.0 made its debut in a rather unusual way: no launch event, no technical blog, no company endorsement, it appeared anonymously on the Artificial Analysis Video Arena leaderboard and reached the top.

Leaderboard data shows that in the text-to-video (without audio) track, HappyHorse-1.0 scored 1379 Elo points, 106 higher than Seedance 2.0, and also higher than Kunlun Wanwei's SkyReels V4 and Kuaishou's Keling AI 3.0 and other mainstream products. In the image-to-video (without audio) track, the model scored 1411, surpassing Seedance 2.0's 1356 and xAI’s grok-imagine-video’s 1331. According to Wallstreetcn, the score in the image-to-video track set a new record for the leaderboard.

It is worth noting that in the comprehensive "video + audio" ranking with native sound effects, HappyHorse ranked second, slightly lower than Seedance 2.0.

Artificial Analysis employs a blind testing mechanism, allowing real users worldwide to compare generated videos without knowing the model origin, and aggregates results through a chess-like Elo rating system. Compared to traditional vendor-reported evaluation methods, this mechanism is considered to provide a more objective signal of model quality. However, due to the relatively limited sample size for new models, their Elo scores tend to fluctuate more than older models; additionally, Elo is a relative score system, and the industry generally believes a gap of about 60 points represents a stable winning advantage.

Controversy and Doubts

HappyHorse's unexpected rise is not without controversy. Some observers have pointed out that over 60% of blind test samples in Artificial Analysis involve human portrait and talking head content, and HappyHorse has a natural advantage in portrait scenarios, which may contribute to discrepancies between evaluation scores and actual comprehensive abilities.

On social platform X, discussants are divided into two camps: skeptics believe that HappyHorse still lags behind Seedance 2.0 in character detail and motion coherence; supporters see great potential, especially hoping it can solve the industry pain point of quality consistency across multi-shot sequences.

ATH Division: Alibaba’s New AI Acceleration Engine

The Alibaba ATH Division behind HappyHorse—full name Alibaba Token Hub Division—was officially established on March 16 this year. According to The Paper, Alibaba CEO Wu Yongming issued an internal announcement declaring the formation of this division, aiming to "create, deliver, and apply Tokens" as its core goal, with Wu Yongming directly in charge.

On April 8, Wu Yongming again issued an internal letter, announcing the establishment of a Technology Committee at the group level, with himself as the team leader and members including Zhou Jingren, Wu Zeming, and Li Feifei. Zhou Jingren serves as Chief AI Architect, Li Feifei heads Alibaba Cloud technology and AI cloud infrastructure, Wu Zeming leads the Group Business Technology Platform and AI inference platform. Meanwhile, Tongyi Lab has been upgraded to the Tongyi Large Model Division, headed by Zhou Jingren.

According to Bloomberg, Alibaba’s previous Wan brand video generator ranked only around 20th place on Artificial Analysis, so HappyHorse's rise to the top represents a significant leap in Alibaba’s video AI capabilities. Alibaba states that 2026 is a key year for accelerating AI development, from the establishment of ATH Division to the latest Qwen 3.6 Plus model taking the crown for weekly global usage on OpenRouter. Recently, Alibaba’s moves in AI continue to emerge.

Currently, video generation is one of the most fiercely competitive areas for AI developers and is among the few proven commercially viable directions. With OpenAI's exit from this field last month, opportunities for Chinese companies have expanded further; most leading products on the Artificial Analysis leaderboard now originate from Chinese firms.

Alibaba ATH states that HappyHorse is part of its plan to explore new interactive modes for the AI era and will launch more products in the future. Opening the API will allow external developers to access the model and further test its commercial potential in real-world scenarios.

Risk Warning and DisclaimerThe market has risks; be cautious when investing. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their particular situation. Investing based on this information is at your own risk.