Alibaba's stealth video model
A stealth video‑generation model from Alibaba surfaced in benchmarking chatter and reportedly topped global leaderboards, signalling fast progress out of China in generative video (x.com). The benchmark attention matters because video generation is one of the highest compute and commercialisation frontiers for multimodal AI, and rapid advances there reshape competitive dynamics. (x.com)
A video model called HappyHorse-1.0 appeared on the Artificial Analysis leaderboard around April 7 without saying who built it, and by April 10 Alibaba confirmed it came from its ATH AI Innovation Unit. The model had already climbed to No. 1 in blind-test rankings before anyone knew the company behind it. (cnbc.com) (artificialanalysis.ai) The leaderboard it topped is not a press release scoreboard. Artificial Analysis ranks models by blind human votes, which means people compare outputs without seeing the brand name first. (artificialanalysis.ai) On April 10, Artificial Analysis showed HappyHorse-1.0 at an Elo score of 1,355 in text-to-video, ahead of ByteDance’s Dreamina Seedance 2.0 at 1,273 and ahead of Google Veo 3 at 1,221. Alibaba’s older Wan 2.6 model sat much lower at 1,188. (artificialanalysis.ai) That gap is why people noticed. A jump of more than 80 Elo points over the nearest listed rival is the difference between edging out competitors and landing in a different tier of quality. (artificialanalysis.ai) Video generation is one of the most expensive corners of artificial intelligence because the model has to draw many images in sequence and keep the motion consistent from frame to frame. Companies then have to serve those jobs fast enough for advertisers, studios, and app users who expect results in minutes, not hours. (cnbc.com) (marketwatch.com) That is also why the field has become a commercial race rather than a science fair. CNBC reported that OpenAI recently discontinued its Sora video generation app and platform as it shifted resources toward coding tools, corporate clients, and artificial general intelligence work, while ByteDance’s Seedance 2.0 rollout was paused amid copyright disputes with major Hollywood studios and streaming platforms. (cnbc.com) Alibaba did not roll this out with a big launch event. A newly created X account said only that the team was part of Alibaba’s ATH AI Innovation Unit and that the project was still under development, and Alibaba later told CNBC the account was genuine. (cnbc.com) (bloomberg.com) The stealth launch set off a guessing game across China’s artificial intelligence scene because people were trying to work backward from the output quality. CNBC reported that speculation centered on whether the model came from a giant like Alibaba or Tencent or from a smaller independent lab. (cnbc.com) Investors treated the reveal as more than internet gossip. Alibaba’s Hong Kong-listed shares closed 2.12% higher on April 10 after its involvement became public, following a 6.75% rise on April 8 during broader technology buying and speculation that Alibaba was the mystery developer. (cnbc.com) Alibaba has been pushing harder into artificial intelligence under Chief Executive Eddie Wu, who has called it the company’s top priority across businesses that include cloud computing, chip design, e-commerce, advertising, and entertainment. A stronger in-house video model gives Alibaba something it can plug directly into those products instead of treating video generation as a side demo. (cnbc.com) The bigger shift is geographic. On the same leaderboard where HappyHorse-1.0 sits first, the top ranks are crowded with Chinese models from ByteDance, Skywork AI, Kling AI, Alibaba, MiniMax, and Vidu, while OpenAI’s Sora 2 Pro sits 20th. That does not settle who will win customers, but it does show that the center of gravity in artificial intelligence video is no longer sitting in one country. (artificialanalysis.ai)