MiniMax unveils H3, a multimodal video AI integrating text and audio... "Challenges Sora with open weights"

MiniMax has launched its multimodal video generation model H3 on the Hailuo platform and via developer API.

H3 accepts text, images, video, and audio together, generating up to 15-second 2K videos with editing, motion transfer, voice, and stereo sound support.

It ranked No. 1 in video editing, No. 2 in text-to-video, and No. 3 in image-to-video on the Artificial Analysis benchmark.

MiniMax also plans to release the model weights under a community license and is positioning H3 at a lower price than rivals.

With an open-weight strategy and design choices that account for Chinese AI chips, it is challenging a market dominated by closed models.