Seems like if want this you need to use a frontier model through GCP or AWS if you really want to avoid training