Alibaba releases Happy Horse video generator, which tops global benchmarks

Alibaba quietly released the Happy Horse video generator, and it immediately took first place in global benchmarks, ahead of Sora, Runway Gen-4 and Kling. For Western AI companies this is an unpleasant surprise: first place in one of the most technically demanding segments now belongs to China.

A quiet market entry

Alibaba made no big announcements: Happy Horse appeared without a press conference or a marketing campaign. The model launched "under stealth", quietly and almost unnoticed. Even so, within a few days it spread through developer communities and made headlines in the technical press once the results of the first independent tests proved convincing.

This approach, minimal noise at release followed by an organic wave, has become a signature style of Chinese AI labs. DeepSeek R1 appeared in January 2025 without press releases, Qwen launched quietly, and Wan Video came out almost unnoticed, yet each time the community built the momentum itself. Happy Horse follows the same logic.

The name apparently refers to 快马 (kuài mǎ), "fast horse" in Chinese. A fitting name for a model that aims to outrun all its competitors.

What the benchmarks say

According to Bloomberg Tech, Happy Horse took first place on the key public video generation benchmarks. The evaluation covered several criteria:

• Physical plausibility of the motion of objects and characters
• Accuracy in following text instructions (text-to-video alignment)
• Character consistency over several seconds
• Visual detail and frame sharpness
• Generation speed at comparable quality

The model beat OpenAI's Sora, Runway Gen-4 and Kuaishou's Kling, the three strongest players, which had previously shared the top spots in global rankings. It is the first time a Chinese AI model has taken the outright top position specifically in video generation, unlike in text (DeepSeek) and multimodality (Qwen), where the lead changed hands more often.

Independent testers in the AI community noted that Happy Horse is especially strong in dynamic scenes: moving water, fabric in the wind and facial expressions, all traditionally hard tasks for video models.

Why video became the main bet

Video generation is the new main front in the AI race. In 2024 the main fight was over text models and image generators, but by 2026 all the major labs had switched to video: the compute barrier is higher, the monetization potential is greater, and the competitive advantage is harder to copy.

Alibaba is no newcomer to this race. The company is actively developing the Qwen series and has shown strong results in multimodal tasks, while its cloud division, Alibaba Cloud, competes with Azure and GCP in enterprise AI infrastructure. Happy Horse is a logical continuation of that strategy, aimed at capturing the market for tools for content creators, studios and advertising agencies.

"Alibaba has given

China the crown in video generation," — Bloomberg Tech, April 2026.

For Western players this means tougher competition. OpenAI and Runway are actively building infrastructure and investing in video generation, but both now sit behind first place.

What this means

First place on the benchmarks is a signal to the whole market: the AI race in video is no longer an exclusively Western story. For businesses it means one more strong, and potentially cheaper, tool in the stack for advertising, training content and video production automation. Alibaba traditionally prices its models competitively, which may change the pricing pressure in the AI Video API segment.