The new rate limit system is completely broken. See https://github.com/orgs/community/discussions/180092 for a lot of affected users.
Some report wait times of 50 or 100 hours. I had the issue that it shows "wait 20 seconds" but (on retry) it counts down one second every 10-20 seconds.
And all this for a service where you pay per use.
and here too. https://github.com/orgs/community/discussions/192435
I've been forced to use the unpredictable "auto" mode that chooses GPT 5.4 nine times out of ten. It's incredibly frustrating that they're saying that what amounts to random selective outages is intended behaviour.
I was given 44 hours. Hopefully enough users are affected that somebody higher up tells them to quit it.