Short answer: HappyHorse 1.0 is no longer a mystery. It appeared anonymously on the Artificial Analysis Video Arena in early April 2026, took the top spot in the no-audio text-to-video and image-to-video rankings, and was confirmed days later as an Alibaba project. Alibaba opened a limited beta on April 28, 2026, and released HappyHorse 1.1 on June 23, 2026, through the HappyHorse website and the Alibaba Cloud Model Studio API. Compared with Veo 3.1, HappyHorse offers longer clips (up to 15 seconds vs 8) and built-in video editing; Veo 3.1 offers a higher resolution ceiling (up to 4K vs 1080p), first-and-last-frame control, and broader availability through Google's apps and APIs.
How the "mystery challenger" was unmasked
The story behind the title moved quickly:
- Around April 7, 2026 — a model named HappyHorse-1.0 appeared on the Artificial Analysis Video Arena with no stated owner. The arena ranks models by blind votes: people see two clips for the same prompt and pick the better one. HappyHorse reached the top Elo rating in text-to-video and image-to-video without audio, as CNBC reported.
- April 10, 2026 — CNBC reported that Alibaba was the creator, with the team describing HappyHorse as a project of its Alibaba Token Hub (ATH) business unit, still under development.
- April 28, 2026 — Alibaba announced a limited beta, with access through the HappyHorse website, Alibaba Cloud Model Studio's API and the Qwen app.
- June 23, 2026 — Alibaba released HappyHorse 1.1 on the HappyHorse website and the Model Studio API.
Alibaba's announcements describe hosted access only — website, API and app — so plan on HappyHorse as a hosted service.
Two cautions about the leaderboard result. Arena rankings shift as models are added and votes accumulate, so a first place in April is a snapshot, not a permanent ranking. And the no-audio categories judge the picture alone; a model's position with audio can differ.
HappyHorse 1.0 vs Veo 3.1: specs
The table covers HappyHorse 1.0 as Alibaba Cloud announced it on April 28, 2026 (see the 1.1 note below); Veo 3.1 figures from Google's Gemini API documentation. Checked September 28, 2026.
| HappyHorse 1.0 (Alibaba) | Veo 3.1 (Google DeepMind) | |
|---|---|---|
| Released | Limited beta, April 28, 2026 | October 15, 2025, updated January 2026 |
| Clip length | Up to 15 seconds | 4, 6 or 8 seconds |
| Resolution | Up to 1080p | 720p, 1080p, 4K |
| Aspect ratios | Not stated in the announcement | 16:9, 9:16 |
| Audio | Synchronized audio, including lip-synced dialogue | Always on: effects, ambience, dialogue |
| Generation modes | Text, image and subject to video | Text and image to video; first and last frame; up to 3 reference images |
| Editing | Video-to-video and subject-and-video-to-video | Extension (+7 s steps at 720p) |
| Multi-shot | Multi-shot sequencing | Prompt-directed cuts within a clip |
| Access | HappyHorse website, Alibaba Cloud Model Studio API, Qwen app | Gemini app, Flow, Gemini API, Vertex AI |
Google describes Veo 3.1's 1080p and 4K output as upscaled; treat 4K as a delivery format rather than a promise of more generated detail.
What changed in HappyHorse 1.1
HappyHorse 1.1 (June 23, 2026) keeps the same basic shape: Alibaba's Model Studio documentation lists text-to-video, first-frame image-to-video and reference-to-video models that generate 3–15 second clips at 480p, 720p or 1080p, 24 fps. Alibaba says 1.1 improves motion, subject consistency, prompt adherence, visual quality and audio over 1.0. The comparison below still holds: longer clips on HappyHorse, a higher resolution ceiling and frame control on Veo 3.1.
Where HappyHorse 1.0 is stronger
- Longer single clips. Fifteen seconds gives room for a small scene with a setup and a payoff, instead of one beat.
- Editing built in. Alibaba lists video-to-video editing, so you can revise an existing clip instead of regenerating from scratch.
- Subject-driven generation and multi-shot sequencing, which Alibaba positions for advertising, e-commerce and short-form video with consistent characters.
- Strong blind-vote results for picture quality at launch, per CNBC's April 10 report.
Where Veo 3.1 is stronger
- Resolution. Veo 3.1 delivers up to 4K (on Veo 3.1 and Veo 3.1 Fast); HappyHorse tops out at 1080p.
- Precise shot control. A start image, an optional end image and up to three reference images let you decide how a shot begins, ends and who appears in it.
- Availability and track record. Veo 3.1 has been available since October 2025 across Google's consumer and developer products, with documented limits and safety rules. HappyHorse is newer: 1.0 started as a limited beta in April 2026, and 1.1 has been available through Alibaba's website and Model Studio API since June 2026.
Which should you use?
Choose HappyHorse if your idea needs one continuous 10–15 second take, or you want to edit clips in place.
Choose Veo 3.1 if you need 1080p or 4K delivery, want to pin a shot with start and end frames or reference images, or need a model that is broadly available today.
Whichever you lean toward, test it on your own prompt. A leaderboard tells you which model wins on average across many voters' prompts; it cannot tell you which one handles your product, your character or your camera move.
Trying Veo 3.1 here
Veo Video runs Veo 3.1, Veo 3.1 Fast and Veo 3.1 Lite. Every generation is an 8-second clip with sound, in 16:9 or 9:16, at 720p, 1080p or 4K, from text or from an image with an optional end frame; Veo 3.1 Fast also takes up to three reference images. Extension and shorter lengths are not available here, and HappyHorse is not offered on this site.
New accounts get 30 credits, enough for one Veo 3.1 Lite clip at 720p. Move to Veo 3.1 for final-quality shots. If your scene needs a longer single take, the AI video generator also runs MiniMax H3 Max.
More comparisons: Veo 3.1 vs Kling AI and Veo 3.1 vs Sora 2.