I built a cross-API rate limit reference because I was tired of guessing
I kept hitting 429s on APIs I use daily. Every time, I'd search "Stripe rate limit" or "OpenAI rate limit" and land on a docs page that told me about one API, in one tier, with no error shape and no retry logic. So I st
I kept hitting 429s on APIs I use daily. Every time, I'd search "Stripe rate limit" or "OpenAI rate limit" and land on a docs page that told me about one API, in one tier, with no error shape and no retry logic.
So I started writing down what I actually needed:
- The RPM and TPM by usage tier
- The exact JSON the 429 returns
- Retry logic that actually reads the
Retry-Afterheader - One escape hatch if I'm stuck
The first page covers OpenAI's 2026 limits — Build / Launch / Grow tiers, the difference between slow_down (429) and server_is_overloaded (503), and a Python retry wrapper that respects Retry-After instead of blindly backing off. Every number is dated.
It's here if it's useful: https://api-ceilings.github.io/theceiling/
No signup. No tracking. Just the numbers and the sources. I'll be adding more APIs one at a time — if there's a specific API whose limits you keep hitting, tell me and I'll prioritize it.
Originally published by Dev.to WebDev. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.