MiniMaxLaunchAI ModelsAugust 28, 2026

MiniMax FastH3 runs locally on Apple Silicon and DGX Spark

FastH3, a 4-step distilled video+audio model, now runs on Apple Silicon via MLX (needs 36GB unified memory) and on NVIDIA DGX Spark via CUDA 13. Two Sparks can generate one clip together. Hao AI Lab published the FastVideo Cookbook for local deployment.

How this story unfolded

5 days · 2 reports · 6 community posts · from Aug 28

  1. Aug 28
  2. Aug 29
  3. Aug 31
  4. Sep 1
  5. Sep 2

More stories today

Open the live feed