MiniMax unveils H3 2K video model costing less than one third

A Chinese artificial intelligence startup has launched a new model capable of generating professional-quality video at a fraction of the current market cost, a move that could reshape the economics of the AI sector and challenge the business models of decentralized AI platforms. MiniMax Group released its H3 multimodal video generation model on July 31, 2026, a system that produces 2K resolution videos with stereo sound for less than one-third the cost per second of mainstream competitors, according to the company.

The H3 model can generate videos up to 15 seconds long at native 2K resolution. It accepts text, images, videos, and audio simultaneously, allowing users to input a combination of elements like a logo, script, reference clip, and soundtrack to receive a coherent video output. MiniMax is positioning the tool for commercial applications such as advertising, branding, e-commerce product videos, and gaming assets, rather than for casual social media creators. The company also plans to release the model's underlying code as open-source in early August, enabling anyone with sufficient hardware to run, fine-tune, and build products on top of H3 without paying licensing fees.