Mystery Model HappyHorse-1.0 Takes Over AI Video Rankings, Dominating with Gap Lead

Mystery model HappyHorse-1.0 recently topped the AI Video Arena rankings by Artificial Analysis, securing first place globally in text-to-video, image-to-video, and audio-included tracks. Its Elo scores far exceed competitors like Seedance 2.0, leading by a wide gap.

Sources say the model comes from Alibaba's Taobao Future Life Lab, led by Zhang Di. It uses a 40-layer single-stream Transformer architecture, natively supports synchronized audio-video generation, outputs 1080P videos in 8 denoising steps, and takes only 38 seconds to generate a 5-second clip on H100 GPU. It supports multiple languages and will be fully open-source.

Currently, HappyHorse-1.0 has no official website. The team is working overtime, and an official release is expected on April 10, which is worth looking forward to.