news
Alibaba's anonymous model took the #1 spot on the video leaderboard. By September, it had lost it too.
HappyHorse-1.0 showed up on Artificial Analysis with no listed owner, climbed to the top of the blind rankings, and was only later confirmed as Alibaba's. Then the leaderboard moved on, same as it always does.
In April, a model called HappyHorse-1.0 appeared on the benchmarking site Artificial Analysis with no company attached to it, and climbed to the top of the blind-tested rankings for both text-to-video and image-to-video generation. Nobody outside the lab that built it knew who was behind it, until a newly created account on X, tied to Alibaba's ATH AI Innovation Unit, claimed it. Alibaba confirmed to reporters that the account, and the model, were genuine.
What it actually is
HappyHorse-1.0 is a 15-billion-parameter model whose stated differentiator is joint generation: most competing models produce silent video and hand audio off to a separate pipeline, where HappyHorse-1.0 generates dialogue, ambient sound and foley in the same pass as the picture. Reported figures put 1080p generation at roughly 38 seconds on a single Nvidia H100, with a short 256p clip taking around 2 seconds. Developer access opened through the platform fal on 27 April, across four endpoints: text-to-video, image-to-video, reference-to-video and video-editing.
The part that matters more than the model
What makes this worth writing about on a benchmark site is not HappyHorse-1.0 specifically. It is how fast it stopped being the story. Later reporting put it at number two on the same leaderboard it had anonymously topped, passed by newer releases as OpenAI's Sora and ByteDance's Seedance also slid down the same table. A model that launched in stealth, got outed as a trillion-dollar company's project, took the top spot, and lost it again, all within about five months, is close to the median lifespan of a leaderboard position in this category right now, not an outlier.
That is the same instability we build our own methodology around: a single snapshot benchmark is a claim about one moment, not a permanent ranking, which is why we retest rather than publish a number once and leave it standing. Our own leaderboard carries the same caveat, and Runway's own model is the other recent example of a leaderboard position that did not survive the year.
Sources: Bloomberg, CNBC, VentureBeat.
Daniel Ochoa: Covers model launches, shutdowns and pricing changes as they happen. Reads deprecation notices for a living so you do not have to.