With HappyHorse 1.1 on Alibaba Cloud, OpenAI's Sora discontinued, and ByteDance's Seedance pulled from global release, the top of the AI video board has reshuffled fast.
2026-06-24 · by Moe Ameen
Alibaba Cloud released HappyHorse 1.1 in late June 2026, an upgrade to the AI video model that first reached the top of the Artificial Analysis Video Arena under an alias in April. According to a VentureBeat report published June 23, Alibaba's HappyHorse sits at No. 2 across the arena's blind, head-to-head rankings for both text-to-video and image-to-video — the boards where users vote on which generated clip better matches a prompt. Alibaba framed 1.1 as a production-focused update, citing smoother motion, steadier face consistency, tighter prompt adherence, and native audio paired with the video. It is live on Alibaba Cloud's Model Studio with full API access, and Alibaba paired the launch with a roughly 40% sitewide discount for the first two weeks.
What makes the No. 2 ranking notable is what happened above and beside it. OpenAI announced on March 24, 2026 that it was discontinuing Sora, its text-to-video product; the web and app experiences were shut down in late April and the API is set to end in September. OpenAI did not give a single official reason, but reporting tied the decision to the cost of running video generation — described as roughly a million dollars a day against a small fraction of that in revenue — after the app's user base spiked at launch and then fell off sharply. Separately, ByteDance shelved the international rollout of Seedance 2.0 amid copyright complaints from Hollywood studios, keeping a leading model off the global market.
The net effect is a reshuffled leaderboard. Two of the highest-profile rivals — one Western, one Chinese — pulled back from the consumer market within months of each other, while Alibaba shipped an upgrade and undercut on price. Treat the exact ranking position, benchmark scores, and discount as a fast-moving snapshot: arena standings shift with every model update, and pricing promotions expire. Confirm current numbers against Alibaba Cloud's listings and the live Artificial Analysis board before quoting them.
The lesson in this reshuffle is to stop betting your workflow on any single video model. In one quarter the field lost Sora to economics and Seedance to copyright, while the No. 1 spot changed hands — that is not a stack you want a content business depending on. Kompozy is built as the layer that absorbs exactly this churn. It generates finished video itself, without riding one external leaderboard model: HeyGen talking-head Persona Shorts with auto-captions, the Persona HeyGen Video Agent for longer multi-scene pieces, a fal.ai VFX hook on Persona VFX HeyGen, avatars composited into brand-exact HyperFrames as Persona Frames, plus Listicle and Marketing Shorts. Your recurring, on-brand format keeps shipping whether or not HappyHorse, Veo, or whatever comes next is ranked first this week.
When you do want to use a top-ranked external model, Kompozy is the finishing and distribution layer that makes its output usable. Generate a clip in HappyHorse through Alibaba Cloud, bring it into Kompozy as a source, and it burns in branded captions, reframes the clip to each destination's aspect ratio, composites it with b-roll or music into a Clipped Short, and fans the same idea into a carousel, a quote card, and platform-native captions in your voice through the Persona Brief — then schedules and publishes the set across all nine connected platforms from one queue. The model supplies interchangeable footage; Kompozy is the durable engine that turns it into a finished, on-brand publishing calendar, and it doubles as the place to cover a story like this — drop your take on the rankings shake-up in as a source and it becomes a blog post, an explainer carousel, and short clips scheduled across your channels.
It is an upgrade to Alibaba's AI video model, released through Alibaba Cloud in late June 2026. Alibaba positioned it as a production-focused update with smoother motion, better face consistency, tighter prompt adherence, and native audio. Alibaba's HappyHorse sits at No. 2 on the Artificial Analysis Video Arena for both text-to-video and image-to-video, and the model is available on Alibaba Cloud Model Studio with full API access.
OpenAI announced on March 24, 2026 that it was discontinuing Sora, its text-to-video product. The web and app experiences shut down in late April 2026 and the API is scheduled to end in September. OpenAI did not give a single official reason, but reporting tied it to the high cost of running video generation against limited revenue after the user base fell off following launch.
ByteDance shelved the international rollout of Seedance 2.0 amid copyright complaints from Hollywood studios, keeping a leading model off the global market. A model that is unavailable in your region effectively drops out of contention regardless of its benchmark score, which is part of why Alibaba's model rose relative to it.
It is risky. Within a few months the field lost Sora to economics and Seedance to copyright while the top rank changed hands. A safer approach is to build on a model-agnostic engine like Kompozy that generates its own persona and avatar video and can finish and publish a clip from any external model, so a single model's discontinuation or pricing change does not break your pipeline.