Stability AI was training 20B and 70B models before it all fell apart, says Mostaque

koltregaskes · x · 2026-09-15

Emad Mostaque adds more detail on Stability AI's collapse: the team had trained the best 4B model and SOTA pretrained models in every other modality, and was in the middle of training its own 20B, 70B and larger models—"and then.. stuff" happened. He reiterates it's pretty sad for a range of reasons. This is a follow-up to his earlier remarks about the UK losing its only frontier lab.

Related event: Stability AI Founder Reveals Details of Company's Decline(2 posts)→

Original post →

More from Companies & People

Companies & People channel →