Is running things in batch mode also cheaper? I don't see it, but it would be a nice way for OpenAI to smooth out their demand