i feel somewhat dumb maybe.
i have an integration with Dall-e-3 mainlyy. I generate Images based on Prompts based on own intellectual property.
when assistant/thread/message became available i created an assistant with tasking and that works fine.
so in my interface i can let my assistant rewrite a prompt to be more ‘something something’ . works fine.
i have spend quite some on this feature testing as i call it. (spend like on images i mean)
if i look at how the costs are divided it is all Imagemodel. no tokens for assistant etc.
i was under the impression that an assistant is connected to a language model so the costs would be tokens.
but i see no tokens spend. am i thinking wrong?
if i run out of credit and i up it, then the assistant starts to work straight away but the images take more time and keep logging that i reached my hard limit. i hope this means that openai treats both different as both think different on my credit level 
thanks
1 Like
No dumb questions! Well, not yours anyway! 
Have you heard of Usage Tiers and rate limits?
https://platform.openai.com/docs/guides/rate-limits/usage-tiers?context=tier-two
Sounds like you’re hitting the DALLE3 cap?
1 Like
thank you and please Let me read into it.
i am one of those who does most via programmatic creation of all sorts of things (as i am my only customer i know , ehmm i think i know what i can hit on the limit but sometimes the limit errors do occur.)
btw my question is not so much on the limits but more on the division of the costs when using 1 api key for multiple purposes.
1 Like
Ah, you’ll really want to handle billing on your own back-end.
That said, API keys have come a long way since they were first introduced. You do have more control on how they’re used, etc.
This page should get you headed in the right direction…
https://help.openai.com/en/articles/9186755-managing-your-work-in-the-api-platform-with-projects
Good luck!