Since OpenAI stoped selling the Pro (20x) plan due to missing CPU capacity, we can be pretty sure they are still “promising” a certain service, while technically they can’t deliver what people paid for.
I do understand the technical limitation when CPU power is missing due to high demand, but then they should stop selling more plans. A grocery store can’t sell products which are not in stock either.
There are now multiple same-day reports of this, so I would consolidate evidence rather than keep hammering retries.
The larger current Community thread is:
There is also an open Codex GitHub issue with current reports of the same “Selected model is at capacity” failure:
If you already have one affected request, the useful fields are:
plan tier;
exact selected model;
surface/app/CLI version;
date/time + time zone;
request ID if already available;
whether all models fail or only particular premium models;
whether the same state follows a fresh chat / another device or is conversation-specific.
I would not deliberately burn more paid attempts just to prove it again. Post the existing details into the main thread or GitHub issue so the current incidents are easier to correlate instead of being scattered across five tiny topics.
Also, I would keep “capacity is definitely caused by X20 sales/CPU shortage” as a hypothesis unless OpenAI confirms it. The observed capacity failures are real; the exact serving/root-cause mechanism is still the part we do not have.