Is Happy Horse publicly available in 2026?

Hey, it's me, Maya. If you've been paying attention to the AI video leaderboards lately, you've already seen the name. Happy Horse — technically HappyHorse-1.0 — appeared on the Artificial Analysis Video Arena in early April 2026, immediately sat at the top of both the text-to-video and image-to-video rankings, and didn't move. That kind of debut gets people asking questions fast.

The problem is that "it ranked number one" and "you can actually use it in your workflow right now" are two different things. A lot of people searching for Happy Horse in April 2026 are running into a mess of third-party sites, conflicting claims, and no clear answer on where to even start. This piece is about sorting that out — what's confirmed, what still needs verification, and what to actually do if you're trying to decide whether to build it into a production workflow.

Why People Are Asking Whether Happy Horse Is Publicly Available

The short version: it debuted anonymously.

Happy Horse first appeared on the Artificial Analysis leaderboard with no public team, no known lab behind it, and GitHub and HuggingFace links on its own site that said "coming soon." That combination — top benchmark performance with no traceable origin — made people pay attention in exactly the way you'd expect. Communities tried to figure out who built it. Theories circulated. Screenshots spread.

The identity question got answered: Happy Horse is Alibaba's model, built under the ATH (Alibaba Token Hub) division, led by Zhang Di, who previously built Kling AI at Kuaishou. A detailed breakdown from WaveSpeedAI documents what was confirmed versus what was still speculative as of early April — worth reading if you want to understand the timeline of how information surfaced.

But knowing who built it and knowing whether you can reliably use it are still different questions. That's what most creators are actually stuck on.

What Access Appears to Be Confirmed So Far

As of the end of April 2026, there are ​two access paths you can treat as verified​.

fal.ai — the clearest current option for creators and developers. On April 26–27, 2026, fal officially launched HappyHorse-1.0 as an API partner. The fal Happy Horse launch explainer covers what the platform tested before going live. You can access the model through four endpoints: text-to-video, image-to-video, reference-to-video, and video-edit. Python and JavaScript SDKs are available. The platform supports 720p and 1080p output, multiple aspect ratios including 9:16 for vertical, and fal guarantees full commercial rights on generated outputs. If you need to start using Happy Horse today without waiting on anything, fal is where that actually works.

The fal HappyHorse-1.0 model page is where you'd get your API key and access the playground. That's the correct entry point — not any of the third-party sites that have appeared under variations of the Happy Horse name.

Alibaba Cloud Bailian — enterprise API testing, with full commercial launch in May. Alibaba opened enterprise-level API testing on April 27, 2026. The official fal launch press release confirms this alongside fal's own launch. If your team is already inside the Alibaba Cloud ecosystem, this is the official route. Full commercial availability is expected sometime in May 2026, with pricing to be disclosed then.

There's also consumer access via the Qwen app, which is lighter and doesn't require API setup.

What Still Needs Verification Before Creators Rely on It

Open-source weights. Alibaba announced that Happy Horse would be released open-source, including base model weights, a distilled model, and inference code. As of April 28, 2026, none of that had appeared on GitHub or HuggingFace. The community tracking repository at github.com/brooks376/Happy-Horse-1.0 is one of the cleaner aggregations of what's actually confirmed versus what's still a promise — the repo notes that its status is "pre-release, awaiting official open-source publication."

This matters for self-hosting workflows. If your use case requires running inference locally or within a private infrastructure, you're still waiting.

Official pricing. Alibaba Cloud Bailian's per-request pricing for the Happy Horse API hasn't been publicly disclosed. fal's own pricing for the endpoints is visible once you're in the dashboard, but Bailian's pricing structure — which would affect any team running production volume through the enterprise route — isn't finalized as of this writing.

The "official website" situation is a mess. There are multiple domains running under the Happy Horse name. Some are third-party platforms that have integrated the API and built their own interfaces around it. Some appear to be marketing sites with claims that don't match what's verifiable. The correct approach: go to fal.ai for API access, go to Alibaba's official channels for enterprise access, and be skeptical of any other site describing itself as the "official" Happy Horse platform until Alibaba makes that clear.

The benchmark numbers — Elo 1333 for text-to-video, topping Seedance 2.0, Kling 3.0, and others — come from blind user preference votes. Those results are real and meaningful. But Elo doesn't measure API uptime, pricing stability, or what happens when you put 50 jobs through the queue on a Tuesday afternoon. It doesn't measure that because it can't. Those are things you find out by actually building on a platform.

Why Unclear Availability Matters for Short-Form Workflows

A tool that's good in a demo but unpredictable in a batch run is a problem you don't want mid-campaign.

If you're making 20 variations of a product clip to test before a launch, or running a weekly faceless content batch, or turning around UGC ad assets on a 48-hour turnaround — you need to know that the thing will be there when you hit generate the fourth time, not just the first. That's a different bar than "the outputs look great in a benchmark."

This format works this week, maybe not next — and that applies to tool access, not just content formats. A model that ranked number one in April and is still in enterprise testing isn't the same as a model with six months of production track record. Right now Happy Horse is somewhere in between.

What Creators Can Do While Waiting

Use currently available alternatives

For short-form content production in April 2026, the models that have proven production-level API access are Kling 3.0, Wan 2.6, and Runway Gen-4.5. Seedance 2.0 is strong but US access is currently restricted due to ByteDance's regulatory situation. These have pricing documentation, usage SLAs, and enough history to know what you're getting.

If you want to test Happy Horse outputs now, fal.ai is the correct entry point. Treat it as a test phase — see how it handles your actual prompts and batch sizes before committing it to a live workflow. Make the first version fast, see if it holds up.

Watch for official updates

The clearest signal that the open-source release and Bailian pricing have landed will come through Alibaba's official ATH channels. Don't rely on third-party aggregator sites to break that news accurately — check Alibaba's official communications directly.

Avoid relying on unverified claims

Multiple sites have appeared using the Happy Horse name with features and pricing structures that don't match what Alibaba has confirmed. Before entering payment information anywhere claiming to offer Happy Horse access, verify that the platform is actually integrated through fal.ai or directly through Alibaba Cloud Bailian. There's no shortcut here: confirm the access path, confirm commercial rights, confirm pricing terms.

Conclusion

Happy Horse is real, it's Alibaba's, and it ranked number one on a benchmark that's hard to game because it uses blind human preference votes. That's a meaningful signal. But "number one on a leaderboard" and "stable enough to run your product launch assets through at scale" are different bars.

Right now, fal.ai is the clearest path for anyone who needs access today. If you're enterprise and already inside Alibaba Cloud, Bailian testing is open. Open-source weights are still pending. And the third-party site landscape is noisy enough that you should verify access paths before relying on anything that isn't directly fal or Alibaba Cloud.

Test it this week if you're curious. Keep your production workflow on proven tools until the pricing and SLA picture is clearer. Come back next month — the open-source release and Bailian commercial launch will likely change the answer.

Go test.