Cutting API costs in OpenAI API

Struggling with API cost. I know about cutting down prompts and being targeted, but is there a more cost effective way to do this? I’m on a complex project and sometimes it takes 10 times for a high level model to get it right.