Development Timeline
From daVinci-MagiHuman to the #1 spot on the AI Video Arena.
daVinci-MagiHuman Open-Sourced
Joint release by GAIR (Shanghai Innovation Institute) and Sand.ai — the 15B-parameter unified Transformer for video+audio generation.
HappyHorse Appears on AI Video Arena
HappyHorse V1 and V2 quietly appeared on Artificial Analysis' blind-test leaderboard with Elo 1333, surpassing Seedance 2.0 and triggering a global "decryption race."
Identity Revealed by Community
X user Vigo Zhao matched HappyHorse benchmarks to daVinci-MagiHuman data point by point. 36kr published a deep investigation confirming the Sand.ai connection.
Full Open-Source Release
Weights, inference code, and model hub expected to go live. GitHub repository (happy-horse/happyhorse-1) currently shows 'Coming Soon.'
Current Status
As of April 8, 2026, Happy Horse is in a high-hype, low-availability state. The model weights, inference code, and model hub are listed as “Coming Soon” on the project website and GitHub repository. The community expects the full open-source release around April 10, 2026. No public API or commercial pricing has been announced.
Once released, the community anticipates a second wave of attention as developers begin benchmarking, fine-tuning, and building on top of the model.