Stability AI AI Updates: July 31, 2026

1. Stability AI Publishes Hardware-Optimized Stable Audio 3 Checkpoints

Stability AI published Stable Audio 3 Optimized on Hugging Face, a set of experimental checkpoints of its text-to-audio latent diffusion model tuned for accelerated inference on specific hardware. The company reports the model can generate several minutes of audio in under two seconds on an NVIDIA H200 GPU or in a few seconds on a MacBook Pro M4, and it ships in small and medium sizes aimed at consumer hardware. The repository supports both text-to-audio generation and inpainting-based editing, and the underlying model was trained on licensed and Creative Commons recordings from AudioSparx and Freesound. Source