MiniMax open-sources multimodal model H3, Moore Threads adapts it in 3 hours

On August 3, MiniMax open-sourced its first multimodal generation model, H3, which accepts text, images, audio, and video inputs to create 2K content up to 15 seconds long. It ranks first globally for video editing on Artificial Analysis and offers video generation at just 0.8 yuan per second, making it commercially viable. The open-source release lowers barriers, allowing enterprises to deploy it locally and customize with their own data.