We have identified the issue and rolled out a fix
Embedding models are returning 4XX errors, we are investigating