Alibaba debuts video AI

Alibaba did not just ship another chatbot. It surfaced as the creator of HappyHorse-1.0, a video model that appeared on the Artificial Analysis leaderboard in April 2026 and quickly took the top spot in text-to-video tests. (artificialanalysis.ai) (cnbc.com) That mattered partly because the model showed up before Alibaba publicly claimed it. CNBC reported on April 10, 2026 that Alibaba confirmed the once-anonymous project was its own after the model had already climbed global rankings. (cnbc.com) A text-to-video model is software that turns a written prompt into a short clip, like typing “a red car drifting through snow” and getting moving footage back instead of a still image. An image-to-video model starts with one picture and animates it into motion, which is useful for product shots, ads, and social clips. (marketwatch.com) (artificialanalysis.ai) The benchmark Alibaba topped is not a company press release scorecard. Artificial Analysis runs a crowdsourced “Video Arena” where models are ranked by blind user preferences, and its text-to-video table now shows HappyHorse-1.0 at 1,355 Elo, ahead of ByteDance Seedance 2.0 at 1,273 and OpenAI Sora 2 Pro at 1,196. (artificialanalysis.ai) The “coming soon” note on that leaderboard is one clue to Alibaba’s real plan. The company is not treating HappyHorse-1.0 as a toy website first; reporting says it is preparing to offer the model to enterprise customers through Alibaba Cloud, its business that sells computing and artificial intelligence tools to other companies. (artificialanalysis.ai) (cnbc.com) That puts the launch in the same lane as Alibaba’s other artificial intelligence push. Alibaba has already been selling models and computing through its cloud arm, and HappyHorse extends that strategy from text and image systems into video, which is harder to build because every second requires many coherent frames instead of one picture. (cnbc.com) (artificialanalysis.ai) The immediate rival is not in Silicon Valley but in China. MarketWatch noted that Alibaba’s move follows ByteDance’s release of Seedance 2.0 earlier in 2026, showing that China’s biggest internet groups are racing to turn video generation into a product category rather than a research demo. (marketwatch.com) That race is getting crowded fast. The same leaderboard places models from Google, OpenAI, KlingAI, PixVerse, MiniMax, and Vidu around HappyHorse, which means Alibaba is entering a field where buyers can compare quality, speed, and price almost side by side. (artificialanalysis.ai) The reason cloud customers care is simple: a retailer can turn one product photo into dozens of moving ads, and a game studio can test cinematic scenes before paying a full animation team. Video generation cuts the cost of making rough drafts, just as image generators cut the cost of mockups two years earlier. (marketwatch.com) (cnbc.com) Alibaba already had a video model family under the Wan name on the same leaderboard, with Wan 2.6 sitting far below HappyHorse-1.0 at 1,188 Elo. That gap suggests Alibaba is not just refreshing an old model but trying to leap into the front rank of video generation with a separate, stronger system. (artificialanalysis.ai) So the story is not only that Alibaba launched a new model on April 10, 2026. It is that one of China’s biggest platform companies used a benchmark win to announce it is ready to sell generative video through its cloud business, turning a flashy research result into a commercial product pitch. (cnbc.com) (marketwatch.com)