I make a lot of daily API calls, and have never had this problem except when I pulled 100s of text strings out of a DB and sent them to the embeddings API without a delay in the loop! That broken things fast, after a few iterations.
However, for all “normal” (not in a fast loop) API call, I have never had this issue, ever.
My best guess is that there is some “strangeness” going on with Cloudflare, based on some access / rate limiting criteria we are not aware of. It is also possible that OpenAI has been so busy that the have not updated their "work flow " to update Cloudflare rate limiting on a “per user” basis. However, I “think” (from all the posts here) the issue resides in Cloudflare.
You can see from this Cloudflare doc, that every user would have to match a criteria and the Cloudflare rate limiting rule updated / created:
I’m not a Cloudflare person (or a big fan of Cloudflare, TBH), but it does not seem “trivial” to set up Cloudflare on a “per user” basis based on a OpenAI plan which is outside of the Cloudflare ecosystem.
It could easily be possible that the OpenAI staff is not aware of what must happen on the Clouldflare end of things to support their new paid plans?