MiniMax releases H3 video model as China races in AI
MiniMax said H3 can make 15-second 2K videos with stereo sound, a fresh sign Chinese AI firms are pushing higher-end generative video.
MiniMax released its H3 video model, a Shanghai-built system the company says can generate clips up to 15 seconds long in 2K resolution with native stereo sound. The launch lands in China’s race with U.S. rivals in generative AI, where video quality, motion realism and prompt control have become a key test of technical depth.
MiniMax described H3 on X as an "omni-reference" model with "commercial-grade generation," "unbeatable cost efficiency" and "open weights." Its API docs call H3 a "new-generation open general-purpose multimodal video model," and the company says the model is available to developers through its video-generation documentation and API pages. Third-party reviews of the documentation say H3 supports 5-to-15-second clips, 24 frames per second, native stereo audio, a 1440p mode and mixed reference inputs.
Those details point to the business case. Video models can be sold into advertising, entertainment, marketing, education and social media, where companies need fast turnaround, consistent style and lower production costs. MiniMax is also signaling that Chinese AI firms are competing on quality and control as well as price, using features such as open weights and reference-based generation to appeal to developers who want more than a novelty clip generator.
The release also fits a wider strategy inside MiniMax. Early testers on social posts were already comparing H3 with other Chinese video systems, including Seedance 2.0, while MiniMax has been expanding beyond video with Hailuo 2.3 and Hailuo 2.3 Fast, Speech 2.8, Music 3.0 and the M3 text model. MiniMax also said on July 8 that it planned to launch a 2.7 trillion parameter model, a parallel bet on frontier text systems that shows the company is building a broader AI stack rather than a single product line.
In a market shaped by export controls, chip constraints and intense geopolitical competition, H3 is more than a feature update. It is another bid by a Chinese startup to prove that domestic models can compete in the most visible, commercially valuable parts of generative AI.
This article was produced by Prism’s automated news system from verified source data, official records, and press releases, then run through automated quality and moderation checks before publishing. The system is built and supervised by the people who set the standards it runs under. Read our full AI policy.
Did this article answer your question?
