HappyHorse-1.0 moved to the top of arena-style text-to-video and image-to-video leaderboards, and creators posted early tests showing strong multi-shot adherence and motion. Its vendor, pricing, and rumored ties to Veo or Hailuo remain unconfirmed, so watch for verification.
You can check the text-to-video leaderboard, the image-to-video leaderboard, and the live Video Arena yourself. The odd bit surfaced fast: the text leaderboard already shows HappyHorse-1.0 at No. 1, while the image-to-video page still lists Dreamina Seedance 2.0 720p on top, even as creator posts were calling HappyHorse the leader in both.
Artificial Analysis gives the cleanest hard datapoint in this story. On its Text to Video leaderboard, HappyHorse-1.0 sits above Dreamina Seedance 2.0 720p, SkyReels V4, and Kling 3.0 1080p.
The site says its rankings come from blind, crowdsourced preferences in the Video Arena, where users submit prompts, watch paired generations, and unlock a personal leaderboard after 30 submissions. Christmas came early for leaderboard obsessives.
The interesting part is not the codename. It is the kind of footage people immediately used to stress it.
Across the first test thread, three patterns keep recurring:
That lines up with what creative users usually notice first in video tools: whether characters, products, and camera logic survive the cut.
The attribution layer is still mostly rumor. Ozan Sihay's post framed HappyHorse as a codename that might point to Google Veo 4, while also mentioning talk around Hailuo 3.0, and another X reply echoed the Veo guess without adding evidence.
The official web evidence is weirder than the rumor mill. Artificial Analysis' image-to-video leaderboard still says Dreamina Seedance 2.0 720p leads that arena, not HappyHorse, so the "No. 1 in both" claim from creator testing has not fully lined up with the public leaderboard page Exa could read.
The only standalone HappyHorse site Exa found was a same-day Terms of Service page. It describes HappyHorseAI as a browser-based service for text-to-video, image-to-video, reference-guided creation, downloads, and credit-based usage, but it does not answer the big question, namely who actually built the model.
