HappyHorse-1.0
#33 in Open-Source LLMsalibaba · v1.0 · since April 2026 · 19× · last seen Jun 30, 2026
15
Momentum
HappyHorse-1.0 is a video generation model by Alibaba (ATH Innovation Unit) that debuted anonymously on the Artificial Analysis platform in April 2026, reaching #1 rank in Text-to-Video and Image-to-Video benchmarks. The model is based on a unified 40-layer Transformer architecture with 15 billion parameters and generates video and audio in a single forward pass without separate audio post-processing. It was officially launched on April 27, 2026, and supports native multilingual lip-sync and various generation modalities (Text-to-Video, Image-to-Video, Subject-to-Video).
Momentum trend
04.04.03.07.
Features
| Key Benchmark (%) | Artificial Analysis Video Arena: Elo 1,333 T2V (no audio, ~+60 Elo vs. Seedance 2.0); Elo 1,392–1,416 I2V (no audio) — Rank 1 in both categories (April 2026) |
| License | Proprietary (closed API model); no public weights; open-source release announced but not yet released (as of April 2026) |
| Multimodality | Text-to-video, image-to-video, reference-to-video (up to 5–9 images), video editing; native audio (dialogue, ambient, foley) in the same forward pass; 7-language lip-sync (EN, ZH, YUE, JA, KO, DE, FR); output: 720p/1080p, 3–15s, aspect ratios 16:9/9:16/1:1/4:3/3:4 |
| Platform | happyhorse.com (end users), Alibaba Cloud Model Studio / Bailian (API), Qwen App (consumer), fal.ai (API partner) |
| Price per 1M Tokens | Pricing model: per second of video output. Alibaba Cloud Model Studio: $0.14/s (720p) to $0.24/s (1080p); fal.ai: $0.14/s (720p) / $0.28/s (1080p) |
| Release Date | April 27, 2026 (limited beta / phased test launch) |