Repository navigation
Codex Desktop unusable: "Selected model is at capacity" for all models on my account #39643
Description
What version of the Codex App are you using (From “About Codex” dialog)?
Version 26.814.41407
What subscription do you have?
ChatGPT Pro 20x
What platform is your computer?
Darwin 25.5.0 arm64 arm
What issue are you seeing?
I am unable to use Codex Desktop on my account.
Codex consistently returns:
"Selected model is at capacity. Please try a different model."
The issue happens for:
- Every conversation
- Every prompt, including simple prompts like "hi"
- Every available model, except GPT-5.4 Mini and GPT-5.3 Codex Spark
- Different reasoning effort settings
Models tested:
- GPT-5.6 Terra
- GPT-5.6 Luna
- GPT-5.5
- GPT-5.4
The error still occurs after switching models.
Important observations:
- ChatGPT works normally with the same account.
- The same account has the same issue on another computer.
- Another user's account works normally on the same computer.
- GPT-5.3-codex-spark worked for one conversation turn, then failed with the same error.
Sometimes Codex starts processing successfully (for example, web search starts), but the final response is replaced by the capacity error.
This looks like an account-specific Codex routing/admission issue rather than a temporary model capacity problem.
What steps can reproduce the bug?
- Open Codex Desktop.
- Sign in with my ChatGPT Pro account.
- Start a new conversation.
- Select any available model.
- Send any prompt (even "hi").
- Wait for response.
- The request fails with:
"Selected model is at capacity. Please try a different model."
- Switching to another model produces the same result.
The issue reproduces consistently across different machines.
What is the expected behavior?
Codex should process the request normally using the selected model.
Changing between available models should allow the request to continue when one model has capacity issues.
Additional information
I have already tested:
- Different Codex models
- Different conversations
- Different prompts
- Different machines
The issue appears isolated to my account because:
- My account fails on multiple devices.
- Other accounts work on the same device.
The same issue happens when I use Chatgpt desktop and Codex cli
Please investigate:
- Account entitlement/status
- Codex backend routing
- Model admission control
- Possible account-level throttling or corrupted state
Screenshot attached showing the repeated error across multiple models.
Activity
yuqi621 commentedon Aug 20, 2026
This issue prevents me from using Codex completely. Please let me know if you need account diagnostic information privately.
github-actions commentedon Aug 20, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
yuqi621 commentedon Aug 20, 2026
btw, I can use chatgpt in website and in ChatGPT desktop normally, the only abnormal is codex
58420LR commentedon Aug 20, 2026
I'm having the same issue today, man. I can use ChatGPT normally o
n the website and the desktop app, but the only thing acting up is Codex.
shaggyrs6-netizen commentedon Aug 27, 2026
Same here
miaomiaoyaoqimiaomaoxian commentedon Sep 6, 2026
Additional Windows report: switching models does not resolve capacity errors
I am also experiencing repeated failures in Codex desktop with this exact message:
During affected periods, every model I try fails, including 5.3. Following the suggestion to switch models does not restore usable access, and this blocks my conversations and work. Requests can sometimes proceed afterward, so this report is about recurring periods of unusability, not a claim that every request always fails.
Environment
OpenAI.Codex, version26.901.6511.0(queried locally; this is the package version, not a version copied from the About dialog).Steps observed
Local diagnostic evidence
A read-only query of local Codex logs confirmed actual
Turn errorentries, not just quoted error text, for these models on September 6, 2026 (UTC):gpt-5.6-solgpt-5.6-lunagpt-5.6-terragpt-6-astraThe failure with 5.3 is my reported observation; its exact model variant and a corresponding log entry were not independently established in this query.
Previously submitted in-app feedback
I already used the in-app feedback command (
/反馈) but did not save the displayed feedback identifier. Local logs recovered the following successful uploads:01a05eb4-6d9d-7cd2-8bae-207d9839f69b019fbc33-ce13-7cf2-be7b-0749dac979f8Both backend log entries state
feedback event uploaded to Sentry,uploaded_attachments=18, andattachments_failed=false. Both corresponding desktopfeedback/uploadcalls completed witherrorCode=null.These are verified thread/session IDs associated with the uploaded feedback, not claimed to be separate Sentry event IDs or RPC request IDs. A separate feedback event ID was not found in the inspected records. Please use the thread IDs and upload times to locate the existing diagnostic bundles.
Expected behavior / request
Switching to an available model should provide a working fallback. If a broader capacity or access condition affects all selectable models, please show actionable guidance rather than repeatedly suggesting a model switch that does not help.
Please investigate the cross-model failures and correlate them with the uploaded feedback. The root cause (capacity, routing, account access, or another condition) has not been established by this report.
This follow-up adds Windows environment details and verified feedback-upload references to the existing issue. No raw conversation contents, credentials, or full local logs are included.
ChicoSir commentedon Sep 8, 2026