From what I know about batch processing/ concurrency in inference this is a pipe dream... Or its going to cost an arm and a leg. I think they're lying or its going to be a much smaller model and not "frontier"