On-chain analysis shows that HappyHorse, Alibaba’s ATH AI team’s video model, surpassed UC Berkeley’s Arena on April 15, achieving a score of 1299 (±21) in video editing. It outperformed xAI’s and Kuaishou’s models by 42 and 48 points, respectively. Arena noted that video editing is a new frontier, with few models currently supporting it. The team is finalizing the 1.0 release in two weeks. Users can test and vote on arena.ai. On-chain data highlights the model’s strong early performance.

ME News reports that on April 15 (UTC+8), according to monitoring by Dongcha Beating, HappyHorse, the AI video model from Alibaba’s ATH AI innovation division, has launched its version 1.0 on the UC Berkeley community blind testing platform Arena (formerly LMArena). In its debut competition, HappyHorse secured first place in the Video Edit category with a score of 1,299 (preliminary score, ±21), outperforming xAI’s grok-imagine-video by 42 points and Kuaishou’s Keling o3-pro by 48 points. Previously, HappyHorse had already topped the text-to-video and image-to-video rankings on the Artificial Analysis evaluation platform; video editing marks its expansion into a new category. Arena’s official team states that video editing is an emerging frontier capability for video models, currently supported by only a few models. The HappyHorse team indicates it is currently in the final optimization phase before official release, with the full version set to launch in two weeks. Community users can now preview and vote on the model at arena.ai. (Source: BlockBeats)