<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/433f24ff-d728-456e-b89b-e6ea9cb7c993.jpeg?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="Image" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/433f24ff-d728-456e-b89b-e6ea9cb7c993.jpeg"/>
A few days ago, a &#34;dark horse&#34; dominated the most influential AI video evaluation platform in the world.
In the early hours of April 8, on Artificial Analysis's Video Arena leaderboard, a video generation model codenamed HappyHorse-1.0 unexpectedly took the top spot. It ranked first in both text-to-video and image-to-video generation, dethroning ByteDance's Seedance 2.0.
For a time, the entire internet was searching for which tech giant was behind this &#34;Happy Horse.&#34;
Soon, the other shoe dropped.
On April 10, according to informed sources, HappyHorse was developed by the ATH team led by Zheng Bo at Alibaba. Wall Street News confirmed with Alibaba that HappyHorse is currently in internal testing and will open its API in the near future.
With the owner identified, some capital began to move in anticipation, and Alibaba's stock price turned green.
The capital market's reaction is not hard to understand.
Just how capable is HappyHorse? On the Artificial Analysis leaderboard, HappyHorse ranked first with 1,365 points, while Seedance 2.0 had 1,273 points. Two models from KeLing ranked fourth and sixth, respectively. In the image-to-video track, HappyHorse led the runner-up by a full 48 points. Meanwhile, the total point difference between second and tenth place was only just over 50 points.
Artificial Analysis's scoring mechanism involves blind testing by thousands of users—they do not know which model is which; they only judge which video generated from the same prompt is better and then cast their votes. Brand filters and leaderboard manipulation are essentially ineffective under this mechanism. The quality of this horse was voted on with real results.
After the mystery was solved, what is truly worth pondering is the origin of this horse.
The name Zheng Bo is not unfamiliar to those who follow Alibaba AI, but his previous label was never &#34;model training.&#34; Since joining Alibaba in 2017, he has successively served as the head of algorithm for Taobao search and recommendation, CTO of Alimama, and head of algorithm technology for Taobao and Tmall Group. His resume is filled with technologies closest to transactions, such as search, recommendation, and advertising.
In other words, Zheng Bo is one of the technical executives in the Alibaba system who understands commercial scenarios the best.
The team that built HappyHorse-1.0 was the former &#34;Future Living Lab&#34; under Taobao and Tmall. In Alibaba's latest round of organizational restructuring, this team was moved into the ATH Business Group, under the AI Innovation Business Unit. This is also why HappyHorse was initially rumored to be the work of Taobao and Tmall.
This background is also very important.
Prior to this, Alibaba's video generation model research and development was mainly led by the Wanxiang team under Tongyi Lab—that was Alibaba's &#34;regular army&#34; in AI, following the path of fundamental model R&amp;D.
The emergence of HappyHorse signifies that the ATH Business Group has cultivated a second team capable of top-tier multimodal model training, and this team's genes naturally carry an understanding of commercial scenarios and user needs.
One lab handles fundamental research, while another team grown from business scenarios handles application innovation—walking on two legs. This is not the simple logic of internal competition (horse racing), but rather a dual-engine structure that Alibaba is intentionally building in the multimodal domain.
The timing of HappyHorse's ascent is also intriguing.
More than a month ago, in the early hours of March 4, Lin Junyang, the former head of Alibaba's Qwen, posted a tweet with only seven words: &#34;me stepping down. bye my beloved qwen.&#34; Subsequently, news surfaced that several core members, including Hui Binyuan, head of Qwen Code, had resigned or transferred positions.
For a time, discussions about &#34;Alibaba's core AI team leaving&#34; and &#34;Qwen losing its soul&#34; were rampant, and the outside world expressed clear concerns about Alibaba's competitiveness in the model domain. Lin Junyang leading the team to take Qwen from obscurity to a global benchmark for open-source models certainly meant the end of an era. But the end of an era does not equate to the loss of combat effectiveness.
Looking at the timeline, Alibaba's adjustments were rapid. Lin Junyang resigned in early March; the ATH Business Group was established on March 16; on April 2, Qwen 3.6 Plus surpassed 1.4 trillion tokens in daily calls on OpenRouter, topping the global charts; and on April 8, HappyHorse unexpectedly took the top spot on Artificial Analysis.
In just one month, Alibaba has simultaneously launched bombshells on both the language model and video model fronts.
Lin Junyang's contribution is undeniable, but when an organization's technical accumulation is deep enough and its talent pipeline is complete, the movement of individual core personnel will not shake the foundation.
Qwen 3.6 Plus proves that the Qwen team can still maintain its iteration pace after Zhou Jingren took over, and HappyHorse proves that there are multimodal teams within the Alibaba system unknown to the outside world, and they already possess the ability to surpass industry leaders.
The significance of HappyHorse is more than just &#34;Alibaba can still fight.&#34; Placing it within the map of Alibaba's entire AI strategy reveals a main thread surfacing: multimodal AI is becoming a direction with increasing weight in Alibaba's AI strategy.
In an internal letter on April 8, Tongyi Lab was upgraded to the Tongyi Large Model Business Unit, with Zhou Jingren appointed as Chief AI Architect. On the same day, HappyHorse topped the video model leaderboard. It is hard to call it a coincidence that these two events happened on the same day.
Alibaba needs to send a clear signal to the market: its deployment in the multimodal domain is not small-scale lab play, but an organized, structured, and multi-front comprehensive offensive.
The Tongyi Large Model Business Unit oversees fundamental model R&amp;D, with the Wanxiang team continuing to advance the underlying technology of video generation under its command; the AI Innovation Business Unit approaches from the application side, with Zheng Bo's team training multimodal models in a way that is closer to actual scenarios.
Both business units belong to ATH, sharing computing resources and data infrastructure while forming differentiation in product direction.
More crucially, Alibaba revealed that HappyHorse-1.0 is just one of the multimodal models independently developed by Zheng Bo's team, and another different multimodal model will be launched soon.
The ATH Innovation Business Unit has launched the &#34;Exploration Plan for New Interaction Methods in the AI Era,&#34; and HappyHorse is part of this exploration direction.
This implies that Alibaba's investment in multimodal AI is a systematic layout. Video generation is just the entry point, and it may be followed by video understanding, multimodal agents, and new forms of human-computer interaction—any of these directions could give birth to the next generation of killer AI products.
From an industry perspective, HappyHorse's ascent has rewritten an existing narrative.
Before this, the competitive landscape in AI video generation was relatively clear:
ByteDance's Seedance series held the top position, Kuaishou's KeLing followed closely, and Sora loomed in the distance across the ocean.
Alibaba had Wanxiang in this track, but its presence was not strong. The implicit consensus in the market was that Alibaba would not secure a large share of the video model market.
HappyHorse has shattered this consensus.
A team grown from the Taobao and Tmall system submitted its model anonymously for blind testing without any pre-heating or promotion and then climbed to the top—this manner of debut is a statement in itself: no reliance on brand halo, no reliance on public opinion, but speaking purely through product strength.
The way it topped the charts is also worth noting.
In dimensions such as multi-camera scheduling, physical motion simulation, and audio-visual synchronization, HappyHorse was rated by blind test users as comprehensively surpassing Seedance 2.0. These are precisely the key capabilities for video generation models to move toward practical application—film production, advertising creativity, and e-commerce content production all require models to be strong enough in these dimensions.
The understanding of user needs accumulated by Zheng Bo's team in search, recommendation, and e-commerce scenarios may be the source of this product strength.
For ByteDance, the emergence of HappyHorse means that Seedance is no longer the undisputed king. Once the perception is established that the ceiling for video models is not in your hands, the subsequent competitive landscape will be redefined.
For the entire industry, HappyHorse proves one thing: in the race for multimodal models, traditional &#34;non-seeded players&#34; can completely come from behind, and the sources of technical breakthroughs are more diverse than imagined.
And this, perhaps, is the true purpose behind Eddie Wu's intensive organizational restructuring over the past month: to shift Alibaba's AI capabilities from being &#34;hero-driven&#34; to &#34;system-driven.&#34;
The happy horse has reached the summit, but the real story has just begun.

BABA

09988

IBUY

- A new video generation model named HappyHorse-1.0 recently topped the Artificial Analysis Video Arena rankings, outperforming ByteDance's Seedance 2.0.  
- Developed by Alibaba's ATH Zheng Bo team, HappyHorse is currently in internal testing and will soon offer an API, triggering a positive market response reflected in Alibaba's rising stock price.  
- This breakthrough indicates Alibaba's commitment to a dual-engine strategy in AI, integrating foundational research with application innovation, and positions HappyHorse as a serious competitor in the video generation landscape.

Wallstreetcn

Alibaba

BABA-W

Amplify Online Retail ETF

KraneShares 2x Long BABA Daily ETF

Global X E-commerce ETF

Hwabao CSI HK Equities Internet ETF

VanEck Retail ETF

Krne Csi China Internet

ProShares Online Retail

BABA 2x Long Daily ETF - GraniteShares

SPDR S&P Retail ETF

Yinhua MSCI China A ETF

E Fund CSI HK Connect Internet ETF

More Cards After HappyHorse? Alibaba Stages a Surprise Counter-Attack in the Multimodal Battlefield