aaaand Polytoken latest just received 0.6.0. now supports model request options for things like fast mode on providers. Specifically supported for OpenAI and DeepInfra for now, but you can extend it as necessary yourself. (also some fixes to support deepseek-v4-flash-0731, it's good!)
PSA - Polytoken stable updating to 0.5.10. Whole bunch of fixes and improvements that I think you'll like, you just get to like them later than the people on the latest track docs.polytoken.dev/changelog/