On April 15, the globally authoritative AI model evaluation platform LM Arena released the latest ranking for the Video Edit (video editing) category. The happyhorse-1.0 model from Alibaba's ATH business group scored 1299 Elo points and ranked first globally, successfully surpassing international mainstream models such as Grok. Notably, this is the first time a Chinese AI team has achieved the top position in this evaluation category, marking the entry of domestic video generation and editing technology into the world's top tier.
LM Arena is recognized by the global AI community as a benchmark for blind testing evaluations. It uses an anonymous comparison voting mechanism, collecting users' genuine preferences for the output of unknown models and generating rankings using the Elo rating system. The happyhorse-1.0 model that made the list belongs to Alibaba's newly established ATH (Alibaba Token Hub) business group, internally known as Alibaba-ATH. This model focuses on high-fidelity portrait expression, natural dynamic control, and precise video editing capabilities, enabling seamless modification and reshaping of video content based on instructions.
Previously, the HappyHorse series of models had also made a strong impression on another authoritative platform, Artificial Analysis's Video Arena ranking. During the anonymous testing phase, the series topped both the text-to-video and image-to-video categories, with Elo scores leading over top domestic and international models such as ByteDance Seedance 2.0, Kuaishou Kelin 3.0, and Google Veo3Fast.
Industry experts believe that the victory of happyhorse-1.0 not only demonstrates Alibaba's fundamental innovation capabilities in the field of multimodal large models but also indicates that AI video technology is evolving from simple "content generation" to "precise editing and controllable production." As Chinese models frequently lead international authoritative rankings, the focus of competition in the global AI video industry is accelerating towards more practical vertical application scenarios.
