MiniMax H3 Video Model Released, Open Source on Aug 3

MiniMax announced the H3 multimodal video model, which will be open-sourced on August 3. It understands text, images, video, and sound, and can generate audiovisual content with dual channels, supporting up to 15 seconds at 2K resolution.

Thanks to technologies like Contextual Omni Representation, MiniMax says H3 is cost-effective, with prices per second at 2K resolution under one-third of mainstream models, and at 768P about half of mainstream 720P models.

Highlights include native multimodal understanding and generation, precise editing and control, and content creation for commercial scenes like film, ads, and e-commerce.