we still see signs that amazon bedrock does not host models properly we see bad model behavior there that we don't observe when going to the native provider it's not just us, some customers using our inference request we do not route there - not really sure what's going on