Something unusual happened in early April 2026. A model nobody had heard of appeared on the Artificial Analysis Video Arena leaderboard — and immediately claimed the top spot.
That model is HappyHorse 1.0.
No press release. No product launch event. Just a sudden #1 ranking in both text-to-video and image-to-video categories, beating every well-funded lab in the process. Here's what makes it worth paying attention to.
What Is HappyHorse 1.0?
HappyHorse 1.0 is an AI video generation model built by Alibaba's ATH AI Innovation Unit — the same team behind Kling AI. It handles text-to-video and image-to-video through a single unified pipeline, and it does something most models don't: it generates video and audio together in one pass.
The architecture is a 15-billion-parameter Transformer with 40 layers. The first and last 4 layers handle modality-specific embedding and decoding, while the middle 32 layers share parameters across text, image, video, and audio tokens — all in one sequence, no cross-attention between separate branches.
The result is a model that produces synchronized dialogue, ambient sound, and Foley effects alongside the visuals, without any post-production audio sync step.
Why the #1 Ranking Matters
The Artificial Analysis Video Arena uses blind pairwise voting. Real users watch two unlabeled videos generated from the same prompt and pick the one they prefer. No brand names. No context. Just the output.
Votes feed into an Elo rating system — the same math used in chess rankings. When users consistently prefer one model's output over another, that model's score rises.
As of mid-April 2026, HappyHorse 1.0 sits at Elo 1381 — a 107-point gap over the second-place model. In practical terms, users prefer HappyHorse's output roughly 65% of the time in head-to-head blind matchups. That lead has actually widened since the model first appeared on the arena.
Here's how the top of the leaderboard looks:
| Rank | Model | Elo |
|---|---|---|
| #1 | HappyHorse 1.0 | 1381 |
| #2 | Seedance 2.0 720p | 1274 |
| #3 | SkyReels V4 | 1243 |
| #4 | Kling 3.0 1080p Pro | 1242 |
| #5 | Kling 3.0 Omni 1080p | 1228 |
A 107-point Elo gap is not a marginal win. It's a decisive one.
Key Capabilities
Native 1080p Output
HappyHorse 1.0 generates video at native 1080p — no upscaling. Clips run 5 to 8 seconds in standard aspect ratios (16:9 and 9:16). The reported inference speed is around 38 seconds per 1080p clip on a single H100.
Joint Audio-Video Generation
Most AI video models produce silent output and require a separate audio step. HappyHorse 1.0 generates dialogue, ambient sound, and Foley effects in the same pass as the video. The model ranks competitively in both the "with audio" and "without audio" categories on the leaderboard.
Multilingual Lip-Sync
The model supports native lip-sync in seven languages: English, Mandarin, Cantonese, Japanese, Korean, German, and French. This is built into the generation process, not added afterward.
Text-to-Video and Image-to-Video
Both modalities run through the same unified pipeline. You can start from a text prompt or upload a reference image — the model handles both without switching between separate specialized models.
Multi-Shot Generation
HappyHorse 1.0 supports multi-shot video sequences with consistent characters and scenes across complex narrative structures. This makes it practical for storytelling and longer-form content, not just single-shot clips.
8-Step Distilled Inference
The model uses a DMD-2 distillation process that reduces denoising to just 8 steps without classifier-free guidance. This is what makes competitive generation speeds possible at the 15B parameter scale.
Open Source
HappyHorse 1.0 is announced as open source with commercial licensing. The weights haven't been publicly released yet, but the team has committed to making them available. For developers and researchers, this means the model will eventually be customizable and deployable without platform lock-in.
Who Built It?
Alibaba confirmed HappyHorse 1.0 as part of its ATH AI Innovation Unit. The model followed the same stealth launch pattern that's become common in the Chinese AI ecosystem — appearing anonymously on the leaderboard before the team publicly claimed it.
For context, Alibaba is also behind Qwen Image 2 and has a track record of releasing competitive open-source models across multiple modalities.
What This Means for AI Video Generation
HappyHorse 1.0 is the first open-source model to reach #1 on the Artificial Analysis Video Arena. Every other top-ranked model is proprietary.
That combination — open source, 1080p native output, joint audio generation, multilingual lip-sync, and a 107-point Elo lead — is genuinely new. It's not a marginal improvement on existing models. It's a different category of result.
Whether the lead holds as more votes come in, and whether the open-source weights deliver on the claimed specs, remains to be seen. But the blind test results are real, and they're not close.
Try HappyHorse AI Video Generation
HappyHorse 1.0 is coming to AI Video Generator Free. Sign up now to get notified when it goes live — you'll receive free credits on launch day and can start generating HappyHorse AI videos immediately.
In the meantime, you can try other top-ranked models including Seedance 2.0 and Kling 3.0 with free sign-up credits.
