Behind HappyHorse: Alibaba’s Stealth AI Video Model That Topped Every Leaderboard

Alibaba just pulled off the AI equivalent of dropping an album with no promo. The Chinese tech giant anonymously released an AI video generation model called HappyHorse-1.0 that quickly climbed to the top of major AI leaderboards, according to The Information. Nobody knew who was behind it until the connection to Alibaba surfaced.

The model didn’t just compete. It dominated. On the Artificial Analysis leaderboard, HappyHorse-1.0 scored 1333-1357 Elo points in text-to-video, beating ByteDance’s Seedance 2.0 by nearly 60 points. In image-to-video, it set a new all-time record at 1391-1406 Elo. That’s not a marginal win. That’s a statement.

What HappyHorse Actually Does

The model packs some genuinely impressive specs under the hood:

  • 15 billion parameters on a unified single-stream Transformer architecture
  • Generates synchronized audio and video in one pass (most competitors handle these separately)
  • Native lip-sync across 7 languages: Mandarin, Cantonese, English, Japanese, Korean, German, and French
  • Produces 1080p cinematic output in 38 seconds on a single H100 GPU
  • Uses 8-step denoising inference with no classifier-free guidance needed

The team behind it reportedly includes Zhang Di, former Vice President at Kuaishou and a technical architect behind Kling AI, one of China’s most capable video generation tools. The team was formerly part of Alibaba’s Taotian Future Life Lab.

Why the Anonymous Drop?

This is a strategic move worth paying attention to. By launching anonymously, Alibaba let the model’s performance speak for itself before brand association could color perception. It’s a tactic we’ve seen in other contexts, but rarely in AI model releases at this scale.

The timing matters too. OpenAI recently shut down Sora, leaving a gap in the AI video generation market. HappyHorse steps right into that vacuum with a model that’s not just competitive but top-ranked.

Open Source and Commercial

What stands out here is the licensing approach. HappyHorse-1.0 is fully open-source with commercial licensing. Full model weights, distilled versions, a super-resolution module, and inference code are all available on GitHub. That’s a significant play against closed-source competitors and gives developers and enterprises direct access to the top-performing video generation model.

Alibaba’s cloud division is reportedly preparing to offer HappyHorse-1.0 to enterprise clients, which would add another revenue stream to its growing AI portfolio. This sits alongside Alibaba’s recent Wan2.7 release, which includes text-to-video, image-to-video, and video editing capabilities at 720p and 1080p.

What This Means for the Market

The AI video generation space just got significantly more competitive. With Sora gone, the field was starting to look like a two-horse race between ByteDance’s Seedance and a handful of Chinese startups. HappyHorse changes that calculus.

For developers and creators, an open-source model that outperforms proprietary alternatives is a big deal. For Alibaba, it’s proof that its AI ambitions extend well beyond language models.

Alibaba stock responded accordingly, with shares rising on the news. More details on the original report are available at The Information.

Scroll to Top