HappyHorse-1.0 is the AI video generation model from Alibaba's ATH (Future Life Lab) unit, part of Taotian Group. It is a 15-billion-parameter unified multimodal model that outputs 1080p video with synchronized lip-synced audio in seven languages. HappyHorse went viral in April 2026 after launching anonymously on the Artificial Analysis benchmark platform and climbing to #1 on both the text-to-video and image-to-video leaderboards. Alibaba confirmed authorship on April 10, 2026, and the public beta launched on April 27 via the official site and Alibaba Cloud Bailian's Model Studio API.

Artificial Analysis runs blind-test Elo rankings where users vote on side-by-side video pairs without knowing which model produced them. HappyHorse-1.0 reached an Elo of 1,357 for text-to-video and 1,415 for image-to-video - the highest scores on either leaderboard at launch. The unified multimodal architecture, 15B parameter scale, and native 1080p output combine to consistently win the blind comparisons against Veo, Kling, and Seedance.

The official HappyHorse API on Alibaba Cloud Bailian is paid (Alibaba ran a 10% early-access discount for enterprise customers at launch). TwoShot gives you free credits on signup so you can try HappyHorse-style video creation without a credit card or a Bailian account. When your free credits run out you can subscribe or buy a credit pack.

Alibaba's ATH (Alibaba Token Hub) AI Innovation Unit, run inside Taotian Group's Future Life Lab. The team launched HappyHorse-1.0 anonymously on April 7, 2026 to gather unbiased benchmark feedback. They revealed Alibaba authorship via a new X account on April 10 once the model topped the leaderboards. ATH is a separate effort from Alibaba's Tongyi Lab (which builds the Wan video series), reflecting Alibaba's two-track AI video strategy.

Sora is no longer available - OpenAI shut it down. While Sora was running, it set the benchmark for AI video physics and led on maximum clip duration (25 seconds). HappyHorse leads on independent blind-test benchmarks (Elo 1,357 / 1,415 on Artificial Analysis) and outputs native 1080p with synchronized 7-language audio out of the box. For practical creation in 2026, HappyHorse is the relevant comparison; Sora is a historical reference.

On the Artificial Analysis arena, HappyHorse-1.0 currently sits above Seedance 2, Veo 3, and Kling 3 in both the text-to-video and image-to-video Elo rankings (no-audio category). Seedance 2 still leads on lip-sync language coverage (8+ vs 7) and multimodal input flexibility (12 reference files). Kling 3 leads on resolution (4K vs 1080p). The honest answer is that HappyHorse wins blind perceptual votes today; the other models lead on specific capability axes. TwoShot lets you try all of them.

Seven languages with phoneme-level lip sync: Mandarin Chinese, Cantonese, English, Japanese, Korean, German, and French. This is fewer than Seedance 2's 8+ but covers the major content markets in Asia, North America, and Europe. The model produces lip-synchronized speech, not just musical or ambient audio.

Both ship out of Alibaba but from different teams. Wan 2.7 is from Tongyi Lab and was released April 6, 2026, focused on a new 'Thinking Mode' for video editing and long-text rendering. HappyHorse-1.0 is from the ATH unit at Taotian Group and was released April 27, 2026, focused on raw generation quality (it tops the blind-test leaderboards). They reflect Alibaba's two-track approach: Tongyi for productivity-focused models, ATH for frontier benchmark performance.

Yes. The official HappyHorse API is available through Alibaba Cloud Bailian (Model Studio) since the April 27, 2026 beta launch. Enterprise tiers include early-access discounts. For developers who want to call HappyHorse-style models without setting up a Bailian account or navigating Chinese cloud onboarding, TwoShot provides assistant-driven access through HTTP endpoints.

Any of the use cases that drove its #1 leaderboard position: cinematic text-to-video shorts, image animation with character consistency, multilingual dialogue scenes (7 languages with lip sync), product showcases, music-video footage, social-format clips. Native 1080p output makes the results suitable for direct upload to YouTube, TikTok, Instagram, and X without upscaling.