Repository navigation
Codex models repeatedly return "Selected model is at capacity" / "servers are currently overloaded" on ChatGPT Pro account with usage remaining #46149
Description
What version of Codex CLI is running?
v0.153.4
What subscription do you have?
ChatGPT Pro x20
Which model were you using?
All
What platform is your computer?
No response
What terminal emulator and version are you using (if applicable)?
No response
Codex doctor report
What issue are you seeing?
Selected model is at capacity. Please try a different model.
Our servers are currently overloaded
What steps can reproduce the bug?
Description
I am experiencing persistent Codex failures on a ChatGPT Pro account even though substantial weekly usage remains available.
The issue reproduces when using the official Codex CLI directly, without any custom proxy or third-party model provider.
Multiple Codex models are affected.
Environment
Plan: ChatGPT Pro
Codex CLI: v0.153.4
Platform: macOS
Authentication: Official ChatGPT account login
Weekly allowance at time of testing: approximately 65% remaining
Reproduction steps
Log in to Codex CLI using the ChatGPT Pro account.
Start a new empty session.
Select one of the affected models.
Send a minimal prompt:
reply only OK
The request fails even though there is plenty of usage allowance remaining.
Results
gpt-5.6-sol
The request fails with:
Selected model is at capacity. Please try a different model.
gpt-5.6-luna
The request also fails with:
Selected model is at capacity. Please try a different model.
gpt-5.6-terra
The WebSocket connection fails first:
Falling back from WebSockets to HTTPS transport.
The HTTPS fallback also fails:
stream disconnected before completion: An error occurred while processing your request.
You can retry your request, or contact us through our help center if the error persists.
OpenAI request IDs:
e66d938c-eea0-4e88-bf60-e865cb95e801
2dbe8a9a-9bee-4896-a313-22c0489cf01f
Additional upstream evidence
The same ChatGPT Pro account is also used through a self-hosted gateway for testing.
The gateway captured the raw upstream error response returned by the OpenAI backend:
Our servers are currently overloaded. Please try again later.
Another OpenAI-generated request ID associated with an upstream failure:
66496861-1782-4847-ae50-39b75ffb420b
The upstream responses observed include:
429
502
503
The gateway is not the root cause because the same failures can be reproduced directly using the official Codex CLI.
Usage status
/status reports that the ChatGPT Pro account still has substantial weekly allowance remaining:
Weekly limit: 65% left
At one point, /status also showed:
GPT-5.3-Codex-Spark
5h limit: 100% left
Weekly limit: 100% left
However, gpt-5.3-codex-spark later disappeared from the model selection list even though its allowance was still shown in /status.
Expected behavior
Codex should process requests normally while usage allowance remains available, or return a clear account/rate-limit message if an account-specific restriction is being applied.
Actual behavior
Multiple models repeatedly return capacity, overload, stream-disconnection, and HTTP 429/502/503 errors.
The behavior appears to be account-specific, model-routing-specific, or related to a backend capacity pool rather than exhausted weekly usage.
Request
Could the Codex team please check the backend traces associated with these OpenAI request IDs?
e66d938c-eea0-4e88-bf60-e865cb95e801
2dbe8a9a-9bee-4896-a313-22c0489cf01f
66496861-1782-4847-ae50-39b75ffb420b
It would be helpful to know whether this account is affected by:
an account-level temporary restriction,
a model entitlement/routing issue,
an overloaded backend pool,
or another Codex service-side issue.
What is the expected behavior?
No response
Additional information
No response
Activity
github-actions commentedon Sep 17, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
server_is_overloadedon GPT-5.6 Sol: one Plus account fails while two other Plus accounts work on same machine/network #46133Powered by Codex Action
CoolEnoughSmile commentedon Sep 17, 2026
#46133 appears particularly similar to my case.
In my case, the affected ChatGPT Pro account also receives:
while substantial weekly usage is still available.
I can reproduce the issue directly with the official Codex CLI across multiple models, and I also captured the same overloaded response from the OpenAI upstream through a gateway.
This makes me suspect the issue may be related to account-specific routing, admission/capacity allocation, or backend pool assignment rather than local networking or exhausted usage quota.
Relevant OpenAI request IDs:
CoolEnoughSmile commentedon Sep 17, 2026
This issue is now causing direct loss of paid subscription time.
I am on a ChatGPT Pro subscription with only a little over 10 days remaining in the current paid billing period. Codex remains unusable across multiple models despite substantial weekly usage allowance remaining.
I also still have two unused banked Codex resets, but using those resets would not address this issue because the failures are caused by backend capacity/service errors rather than exhausted usage limits.
The affected account continues to receive errors including:
and the OpenAI upstream response:
OpenAI request IDs:
Since this is consuming paid Pro subscription time while the service is unavailable, please preserve this incident for any applicable billing adjustment, subscription extension, or refund review.
Additional account evidence:
This further confirms that the failures are not caused by exhausted quota.
czg1024900 commentedon Sep 17, 2026
Same symptom on Windows 10 with Codex CLI
0.154.0-alpha.6.2and Codex Desktop26.908.9136.0, ChatGPT login auth (no API key).Since ~04:00 UTC today,
gpt-5.6-terraintermittently dies mid-turn after 22-76s of streaming, having emitted very few tokens (e.g. 6 output tokens, 0 reasoning tokens):stream disconnected before completion: An error occurred while processing your request. You can retry your request, or contact us through our help center at help.openai.com if the error persists. Please include the request ID <id> in your message.Request IDs observed:
3cfbc752-fe2e-48ef-b51a-ea5bd182642a027a4163-4d10-426c-be5f-89c0e0da853c0e6f1f01-6ae3-4e3c-8a2b-365e65f5c42c85242e91-7f80-4acf-bf89-b60e4bb04a1c2dca0be8-24fb-4fa5-a61f-318d2f03da10f6b52f1-e6e2-4a64-8bd0-39d283afe725Things I checked, in case they narrow it down:
gpt-6-astra,gpt-5.6-sol,gpt-5.6-luna,gpt-5.5andgpt-reserveall return "Selected model is at capacity. Please try a different model." immediately, whilegpt-5.6-terrasometimes succeeds.codex doctorreports 20 ok / 0 fail; auth, config and the Responses WebSocket handshake (HTTP 101) are all healthy.supports_websockets = false).used_percent0.0, secondary 12.0%), so this does not look like a quota issue.