HappyHorse 1.0: The Chinese Model That Beat Seedance — And Launches the Day After Sora Dies

What is HappyHorse 1.0?

HappyHorse 1.0 is a text-to-video and image-to-video model from Alibaba. It was first spotted as an anonymous entry on the Artificial Analysis Video Arena in early April 2026 under the codename "happyhorse" — the same stealth-launch pattern Anthropic used with Claude on the LMSYS Chatbot Arena. By April 10, the model had climbed to #1 on both the text-to-video and image-to-video boards, and Alibaba confirmed authorship shortly after. The team is most likely Tongyi Lab, the group that shipped the open-source Wan 2.1 and Wan 2.2 video models in 2025.

The model is API-only at launch — there is no consumer chat interface like Sora had. It will go live as a paid endpoint on fal.ai on April 30, 2026. fal.ai's standard premium video pricing runs $0.30–$1.20 per generated second; Alibaba has not yet confirmed final HappyHorse rates. There is no announced free tier, no waitlist, and no invite system — the model launches as a paid endpoint and you pay per generation.

The benchmark numbers: HappyHorse vs everything else (Apr 10, 2026)

The Artificial Analysis Video Arena ranks video AI models by Elo — humans pick the better of two side-by-side outputs and the score updates after each vote. As of April 10, 2026, here is how the top of the board looked:

Source: Artificial Analysis Video Arena, snapshot April 10, 2026. Elo updates daily as voting continues.

Two things stand out. First, HappyHorse beats Seedance 2.0 by 47 Elo on T2V and 61 Elo on I2V. For context, those are the biggest #1-vs-#2 gaps the leaderboard has ever recorded — previous chart-toppers like Veo 3.1 and Seedance 1.0 won by 8–15 Elo. In Elo terms, a 50-point gap means the higher-ranked model wins roughly 57% of head-to-head votes; a 60-point gap is closer to 58%. That is a meaningful, not marginal, lead.

Second, Sora 2 sits in 5th place. OpenAI's flagship video model is now beaten on benchmarks by Alibaba, ByteDance, Google, and Kuaishou. That ranking, plus the pending shutdown on April 26, explains why OpenAI is choosing to wind Sora down rather than keep paying GPU costs to run a model that is no longer competitive.

Why the timing matters: the 4-day Sora gap

Sora's consumer app shuts down on April 26, 2026. HappyHorse 1.0's API launches on April 30, 2026. That leaves a four-day window where roughly 500,000 ex-Sora users are looking for a replacement and the strongest new option isn't live yet. If you used Sora for production work, you have three choices for those four days:

1. Bridge with Veo 3.1. Highest accessible benchmark score (1,209 T2V Elo) and the only top-5 model with native audio. Costs $19.99/month via Gemini Advanced. Best if your Sora workflow involved cinematic shots with sound.
2. Bridge with Kling 3.0. Cheapest at $5.99/month with a 66-credit/day free tier. Outputs up to 3-minute clips at 4K. Best if you generated narrative or social clips on Sora.
3. Skip text-to-video, stay productive on talking-head. If your Sora work was spokespeople and explainers (the most common ex-Sora workflow), HeyGen Avatar V ($24/mo annual) is already better than what Sora was doing for that use case — 0.840 face similarity, 200+ avatars, 175+ languages. Free tier available with no credit card.

Then on April 30, switch your text-to-video API integration to HappyHorse on fal.ai. The full migration playbook (export your Sora archive, match your workflow to a replacement, switch before April 26) is in our Sora migration guide.

What you can do right now (HappyHorse isn't accessible until April 30)

HappyHorse 1.0 isn't a tool you can use today — it is a benchmark result and a launch date. Here is what to actually do between now and April 30, 2026, organized by use case:

If you are an API user planning to integrate HappyHorse on April 30, do two things now: (1) sign up for a fal.ai account and complete billing setup so you can hit the endpoint the moment it goes live, and (2) keep your existing Sora API integration running until September 24, 2026 (the API stays live five months past the consumer shutdown) as a fallback while you test HappyHorse outputs against your prompt library.

What changes when HappyHorse goes live on April 30

Once HappyHorse is accessible on fal.ai, the AI video market shape changes in three ways:

• Best-in-class text-to-video moves to API only. Until now, the highest-quality option (Veo 3.1) was bundled into Gemini Advanced for consumers. HappyHorse is API-first — meaning developers and agencies get the top model, but solo creators without engineering chops will still default to Veo 3.1 or Kling 3.0 inside their respective apps.
• Pricing pressure on Seedance 2.0 and Veo 3.1. ByteDance currently charges roughly $0.50–$0.90/second for Seedance 2.0 on fal.ai. If Alibaba prices HappyHorse below that — possible given its 47-Elo lead — expect competitor price cuts within weeks.
• The Chinese AI video lead becomes structural. The top three Elo positions (HappyHorse, Seedance 2.0, Kling 3.0) are now all from Chinese labs. Veo 3.1 is the only Western model in the top 4. That mirrors what happened in open-source LLMs in 2025 with Qwen and DeepSeek.

The editor verdict: if you are migrating from Sora and you have engineering resources, plan your stack around HappyHorse on fal.ai for text-to-video, HeyGen for talking-head, and Fliki for high-volume narrated content. If you don't have engineering resources, default to Veo 3.1 (in Gemini Advanced) and HeyGen, and revisit HappyHorse once a hosted UI appears on top of the fal.ai endpoint.

Related reading

• Sora migration guide: 7 best alternatives ranked — full export-and-switch playbook for the April 26 shutdown.
• Kling vs Veo vs Runway: cinematic AI video head-to-head — deeper dive on the three accessible alternatives.
• Best AI video tools 2026 (top 10 ranked) — the full landscape with hands-on test scores.
• HeyGen Avatar V guide: what's new — the talking-head model that already replaces Sora's spokesperson use case.