Stale capacity banner: "Selected model is at capacity" shown while model continues responding
Open 💬 14 comments Opened Feb 12, 2026 by aexis-b
💡 Likely answer: A maintainer (etraut-openai, contributor)
responded on this thread — see the highlighted reply below.
What version of the Codex App are you using (From “About Codex” dialog)?
260212.1136 (650)
What subscription do you have?
Business (with prepaid usage/credits at time of issue)
What issue are you seeing?
Summary
Codex Desktop shows a capacity error banner for the selected model, but the same model continues to respond normally in the same thread.
Banner text shown (screenshot attached)
Selected model is at capacity
This model is at capacity right now. Try again after Feb 16, 2026, 12:43 PM, or start a new conversation with another model.
Behavior
- Selected model: GPT-5.3-Codex
- Messages and responses continue successfully
- No interruption in generation
- Credits/usage are available (not exhausted)
Frequency
Always (even after logout)
Impact
Confusing and misleading state; suggests outage/capacity block when service is actually available.
Environment
- OS: macOS Tahoe 26.2 (25C56)
- Codex app version: 260212.1136 (650)
- Local time + timezone when observed: Feb 12, 2026, 2:35 PM CST
Screenshot of in-thread capacity banner while responses still succeed
<img width="777" height="331" alt="Image" src="https://github.com/user-attachments/assets/0398ef6c-22d8-453b-b6a1-e6ea400b1f36" />
What steps can reproduce the bug?
- Open Codex Desktop.
- Select model: GPT-5.3-Codex (Low).
- Continue chatting in an existing thread.
- Observe capacity banner appears above composer/thread.
- Send a message anyway.
- Model responds normally despite banner.
What is the expected behavior?
- If the model is truly at capacity, requests should be blocked consistently.
- If requests are succeeding, the capacity banner should not be shown.
Additional information
_No response_
14 Comments
I'm hitting this too. It really got my hackles up because I know I still have over 200 credits remaining
However, seeing as I still seem able to use codex in my VS Code app with the "nag", but have just now downloaded Codex Desktop -- maybe this is all sneaky a ploy to prompt humans to perform this user behavior
Update
After backing out of the new 5.2-codex chat session, archiving and starting another 5.2-codex session, I noticed there was no "selected model is at capacity" message presented, however after half a dozen or so prompts it re-appeared
However
So it's a cosmetic issue and an annoyance, but everything seems to be "ok"
Having the same "issue", happy that its still responding though 🙃
Question: How can I see if my credits are used, and how many?
(Business Plan user, credits bought by admin)
Codex told me:
I hit the $20 Pro weekly limit in the Codex Mac app and was prompted to upgrade or purchase 1,000 credits. I purchased the credits, but immediately received the banner: “Selected model is at capacity”
<img width="480" height="93" alt="Image" src="https://github.com/user-attachments/assets/7b88064a-f691-4dbf-8937-7459f65e1ee5" />
The message says it will reset at 5:02 PM—the exact time my weekly limit resets—which suggests this may be tied to the weekly cap rather than true model capacity.
The model continued responding despite the banner, so this appears to be a stale or incorrect capacity warning.
@eliasws you can check your Codex usage and credit balance by logging in here:
https://chatgpt.com/codex?openaicom_referred=true
I get the banner as soon as I'm transferred from subscription usage to credit usage. It remains there indefinitely until I'm back on subscription usage. I'm using Codex 5.3.
<img width="1056" height="196" alt="Image" src="https://github.com/user-attachments/assets/1a8a8be7-6c9c-4ab5-b003-f29fd20bbd89" />
Pro, 5.3-high, seeing that for the first time.
I don't even understand the concept — what
capacitycodex is talking about? Aren't tokens spendings depend on reasoning level already?Hit this today. Then Codex applet in VS Code has ceased to let me select any model above 5.2.
Looks like "capacity" here applies not to my account (i'm pretty far from both weekly and 5h limits, and also have some credits available), but to the server-side. Are we DOSing Codex?
I just got this in the CLI, it stopped execution and i just told Codex to continue and it just did
Something wrong is happening with the limits policy at OpenAI, seeing "Selected model is at capacity. Please try a different model." at every request thus throwing away 5h limits for resubmitting my requests.
Looks like you're really going the wrong way overlooking at Claude.
@nos1609 that is not an account level error, that's a service capacity issue. I hit a couple of those last night, it just means there's too many active users hitting the model. "Wait a few minutes and resend your prompt" is all that means
@vanschroeder Just loving more Kimi, Grok and Google's approach downgrading the model for free users leaving enough "capacity" room for subscription owners. Kinda more balanced, and You really get what is paid for and the idea behind that subscription model.
Here, on the other side OpenAi's approach looks like I'm paying for being put on hold without any limits being reset, especially considering the fact it just stops mid-turn, and You have to re-iterate the previous request.
@nos1609 what you're not understanding is that they aren't downgrading you. The request gets dropped. Just resend the prompt, it's not hard. If you don't like ChatGPT dropping the request and not charging you when the model is overcrowded please, just leave and use another service. It sounds like grok might be a good fit for you
Just following up that after several Codex App updates, today I got the same "out of credits" message.
Please resolve this issue, it's a minor issue that has been allowed to persist for a long time. It's not a good look and one that can negatively affect user experience and cause users to lose hours, even days of productivity needlessly.
@webfreshener-agent, this thread is unrelated to usage accounting or reporting. I'm going to hide your post to avoid confusing this thread, which is about insufficient capacity on the server for a given model.