StorySoftware

MiniMax publishes H3 video model weights, but its licence excludes the US, EU, UK and South Korea

MiniMaxHailuo video models (H3)

MiniMax put H3's weights on Hugging Face in two variants, one for text and first/last-frame prompts and one for mixing many reference files. Unlike its M3 language model, the licence grants no rights in four jurisdictions, and firms above $20M in yearly revenue need written permission.

  • Base model: 33B-parameter dense transformer, with Qwen3-VL-32B reading the prompt.
  • Full-precision weights 123.6 GB; smallest quantized version 42.5 GB.
  • Ranked first in video editing and second in text-to-video on Artificial Analysis.
Read the original · Press

More on Hailuo video models (H3)

Primer