After HappyHorse, are there more cards to play? Alibaba delivers a "counterattack" on the multimodal battlefield.
```
A few days ago, a "dark horse" dominated the world’s most influential AI video evaluation platform.
In the early morning of April 8, a video generation model codenamed HappyHorse-1.0 suddenly topped the Video Arena leaderboard on Artificial Analysis. It took first place in both text-to-video and image-to-video, dethroning ByteDance’s Seedance2.0.
For a time, the entire internet was speculating which big company created this "Happy Horse."
Soon, the truth came out.
On April 10, sources revealed that HappyHorse is a product developed by Alibaba’s ATH team led by Zheng Bo. Wallstreetcn confirmed with Alibaba that HappyHorse is currently in internal testing and the API will be opened soon.
Once it was clear who the developer was, some funds began scrambling, and Alibaba’s stock price rose.
The reaction from the capital markets is not hard to understand.
How strong is HappyHorse? On the Artificial Analysis leaderboard, HappyHorse ranks first with 1,365 points, Seedance 2.0 has 1,273 points, and two models from Kuaishou's "Kelng" rank fourth and sixth. In the image-to-video track, HappyHorse leads the second place by 48 points, while the gap from second to tenth is just over 50 points.
Artificial Analysis’s scoring mechanism is a blind test involving thousands of users—not knowing which model generated which video, they simply rate which one is better for the same prompt. Brand filters and vote manipulation are basically invalid under this mechanism. The quality of this horse is truly voter-driven.
After the mystery was revealed, the truly interesting thing was the origin of this horse.
Zheng Bo’s name is not unfamiliar to those following Alibaba AI, but he has never been labeled a "model trainer." Since joining Alibaba in 2017, he served as head of Taobao search and recommendation algorithms, CTO of Alimama, and head of algorithm technology for Taotian Group—his resume is all about search, recommendation, and advertising, the technologies closest to transactions.
In other words, Zheng Bo is one of the technical executives in the Alibaba system who best understands business scenarios.
The team behind HappyHorse-1.0 is Taotian’s former "Future Life Lab." In Alibaba’s latest organizational adjustment, the team was integrated into the ATH business group and reports to the AI Innovation division. That’s also why there were initially rumors that HappyHorse originated in Taotian.
This background is very important.
Previously, development of Alibaba's video generation models was mainly led by the Wanxiang team under Tongyi Lab—Alibaba AI’s "regular army" focused on foundational models.
The appearance of HappyHorse means the ATH group now has a second team capable of top-tier multimodal model training, and this team is naturally attuned to business scenarios and user needs.
One lab does basic research; a team emerging from real business does application innovation—the "two legs" approach. This is not simply internal competition but a dual-engine structure that Alibaba is deliberately constructing in the multimodal field.
The timing of HappyHorse’s rise is also intriguing.
Just over a month ago, in the early morning of March 4, Qwen’s previous head Lin Junyang tweeted simply: "me stepping down. bye my beloved qwen." Afterwards, multiple core members such as Qwen Code’s chief Huibin Yuan either left or changed positions.
For a while, speculation was rampant: "Alibaba AI’s core team is leaving," "Qwen loses its soul." There were obvious concerns about Alibaba’s competitiveness in model development. Lin Junyang’s leadership made Qwen a global open-source benchmark, so his departure did mark the end of an era. But the end of an era does not mean the end of combat power.
On the timeline, Alibaba adjusted very quickly. Lin Junyang left in early March, the ATH group was established March 16, Qwen 3.6 Plus on OpenRouter hit a single-day invocation of 1.4 trillion tokens on April 2, and HappyHorse topped the Artificial Analysis ranking April 8.
In just a month, Alibaba played a trump card in both language and video models.
Lin Junyang’s contribution is undeniable, but when an organization’s technical depth and talent pool are complete, the departure of key individuals won’t shake the foundation.
Qwen 3.6 Plus proved the Qwen team can maintain iteration pace under Zhou Jingren’s leadership, and HappyHorse proved there are more, unexplored multimodal teams inside Alibaba, ones already surpassing industry leaders.
The significance of HappyHorse goes beyond "Alibaba is still strong." Placed in the context of Alibaba’s AI strategy, a central theme emerges: multimodality is now a direction of increasing strategic importance.
In the internal memo of April 8, Tongyi Lab was upgraded to the Big Model division, with Zhou Jingren as Chief AI Architect. On the same day, HappyHorse topped the video model leaderboard. Both events on the same day—it’s hard to call this coincidence.
Alibaba needs to deliver a clear message to the market: its multimodal roadmap is not dabbling in a lab, but an organized, full-scale, multi-pronged campaign.
The Tongyi Big Model division oversees foundational research, with the Wanxiang team continuing video generation tech; the AI Innovation division, led by Zheng Bo, trains multimodal models in more business-grounded scenarios.
Both divisions belong to ATH, sharing compute power and data infrastructure, while differentiating in product direction.
More importantly, Alibaba revealed that HappyHorse-1.0 is only one of several multimodal models independently developed by Zheng Bo’s team—another different model will be launched soon.
The ATH Innovation division has launched an "AI-era new interaction exploration plan," of which HappyHorse is a part.
This means Alibaba’s investment in multimodality is a system-level strategy. Video generation is just the entry point. Next could be video understanding, multimodal agents, or new forms of human-computer interaction—any of these fields might produce the next killer AI app.
From an industry perspective, HappyHorse’s rise rewrote the established narrative.
Before this, the AI video generation arena had a clear competitive pattern:
ByteDance’s Seedance series firmly sat at the top, Kuaishou’s Kelng close behind, and Sora looming from across the ocean.
Alibaba had Wanxiang in the race but wasn’t prominent. The market’s implicit consensus was that Alibaba would get only a small share of the video model pie.
HappyHorse broke that consensus.
A team growing out of the Taotian system, with no fanfare or promotion, anonymously submitted a model to the blind test, and then took first place—this kind of debut sends a message: No need for brand aura, no need for hype—let the product speak for itself.
The way it topped the list is also noteworthy.
In aspects like multi-shot scheduling, physical movement simulation, and audio-video synchronization, HappyHorse was rated by blind testers as comprehensively surpassing Seedance 2.0. These are exactly the core capabilities needed for real-world video generation applications—film production, creative advertising, e-commerce content, all need strong performance here.
The Zheng Bo team’s understanding of user needs in search, recommendation, and e-commerce likely powered such product strength.
For ByteDance, HappyHorse’s emergence means Seedance is no longer the undisputed king. Once the industry believes the video model ceiling isn’t in your hands, the competitive pattern is redefined.
For the whole industry, HappyHorse proves: in the race of multimodal models, "non-seeded players" can absolutely outperform, and technological breakthroughs come from more diverse sources than expected.
This might be the real purpose behind Wu Yongming’s intensive organizational adjustments over the past month: to shift Alibaba’s AI power from "hero-driven" to "system-driven."
The happy horse has climbed to the summit, but the real story is just beginning.
Risk Disclosure and DisclaimerThe market involves risk; investments require caution. This article does not constitute personal investment advice, nor does it consider the specific investment goals, financial situation, or needs of any individual user. Users should consider whether any opinions, views, or conclusions in this article fit their specific situation. Investment is at your own risk. ```