MiniMax launches H3 video model in AI race with ByteDance

MiniMax released H3 on Friday, a new video model that can process text, images, video and audio, widening the fight for dominance in China’s fast-growing AI video market. The model can generate clips up to 15 seconds long in 2K resolution with native stereo sound, a package that makes it more than a text-to-video demo and puts it directly into the same commercial lane as tools aimed at marketers, creators and enterprise teams.

MiniMax has positioned H3 as an open, general-purpose multimodal video model and said it planned to release the weights within days, a sign that the company is leaning on openness as much as raw performance. Its product pages describe the system as breaking boundaries between tasks and modalities, and the company has said it wants to support the open-source community and a broad range of hardware. That approach matters in a market where access, portability and speed of deployment can be as important as photorealism.

The launch also sharpens competition with ByteDance and Kuaishou, the two rivals named in the market MiniMax is entering. One search result summary said H3 outperformed ByteDance’s Seedance 2.0 in some tasks, a claim that reinforces how quickly Chinese developers are iterating in video generation and video understanding. MiniMax’s move comes after it said earlier this month that it planned a giant 2.7 trillion-parameter model, showing the company is building both at frontier scale and in smaller products that can be shipped to users faster.

The policy stakes are rising with the technical progress. A model that can stitch together video and synchronized sound from prompts or mixed media raises the same concerns now following the broader generative-AI industry: misinformation, impersonation, copyright-sensitive remixing and the computing power needed to build and run such systems. Those issues are especially sensitive in China, where officials have encouraged AI development while tightening controls around content and data security.

The market reaction suggested investors are watching that trade-off closely, with MiniMax shares jumping after the unveiling. For creators and businesses, H3 points to cheaper and faster video production workflows. For regulators, it is another reminder that the video-AI race is no longer about text chatbots or research prototypes, but about tools that can be deployed at scale.