MiniMax, a China-based AI startup, launched H3 on July 31, calling it a multimodal generation model that takes text, images, video, and audio as input and can generate up to 15-second 2K videos with stereo sound.

The company said it planned to release the model weights in the coming days, subject to applicable laws and regulations.

MiniMax said H3 was aimed at advertising, ecommerce, product design, user interface and user experience design, and gaming.

It also highlighted generation and editing features, including video-to-video motion transfer.

The company said H3 cost less than one-third as much as mainstream models at 2K, and less than half as much at 768p, compared with mainstream models at 720p. But it did not name the comparison models.

MiniMax is among China’s larger AI startups. It reached a US$4 billion valuation in 2025 after a US$300 million series B extension.

The planned weight release comes as Chinese AI companies release more open-weight models.

China’s deep-synthesis rules require labels on some AI-generated content that could mislead the public.

🔗 Source: MiniMax

Recent MiniMax developments