HyperFlow H3 Flash

About

A compressed, few-step version of MiniMax-H3 that generates video and synchronized stereo audio together. It has 13.7B parameters, samples in 4, 6 or 8 steps without classifier-free guidance, and ships GGUF and INT8 weights so it can run on 24 GB consumer GPUs as well as data-center hardware. The README reports a 5.2 second clip in 8.7 seconds on 4 B200 GPUs.