Repository navigation
Account-specific server_overloaded / capacity on Pro despite usage remaining (thread 01a1337b-ddfb-7013-ba44-2716c02d2789) #53555
Description
Summary
Repeated account-specific Codex failures with:
Selected model is at capacity. Please try a different model.
codex_error_info: server_overloaded
This is not a usage-limit block. Weekly window ~20% used / 80% remaining, credits available, rate_limit_reached_type: null. A different ChatGPT account on the same machine and egress works during the same periods. Failures are frequent in daytime Asia/Shanghai; evenings often work on the affected account.
Uploaded thread (Feedback): 01a1337b-ddfb-7013-ba44-2716c02d2789
What version of Codex CLI is running?
codex-cli 0.159.2
What subscription do you have?
ChatGPT Pro (Philippines billing/region context). Account id (for Support correlation): 769e217f-9f0d-4ba9-9a96-4d5bc55494e2
Which model were you using?
gpt-6-astra with reasoning_effort=ultra (also reproduced earlier on gpt-5.6-sol + medium with Fast off). Fast/Ultrafast was off for the 5.6 repro.
What platform is your computer?
Windows 10/11 x64 (PowerShell). Client: Local Codex via ChatGPT auth app-server (not Codex Cloud).
What terminal emulator and version are you using (if applicable)?
Windows Terminal / PowerShell. Also Feishu bots via cc-connect → Local app-server (stdio://).
Steps to reproduce
- Sign in with the affected ChatGPT Pro account (
auth_mode=chatgpt,CODEX_HOME= dedicated home). - Select
gpt-6-astra(ultra) orgpt-5.6-sol(medium), Fast off. - Start Local Codex (CLI or app-server) and send a normal coding prompt (or even short prompts).
- Observe immediate or mid-turn failure:
Selected model is at capacity/server_overloaded. - Control: switch to a different ChatGPT account on the same machine/network — requests succeed.
Expected behavior
When usage/credits remain available, Local Codex should admit the selected model. If overloaded, distinguish account admission/routing from generic capacity and retry/queue rather than terminal failure.
Additional context
- Client egress via local proxy
127.0.0.1:10808→ public IP65.49.203.202(US / California / Los Angeles). codex doctor --json: auth chatgpt ok; websocket handshake HTTP 101 ok; provider reachable; overallStatus warning (Defender exclusions unverified only).- Sample failing turn IDs (UTC+8 = Asia/Shanghai, date 2026-10-13):
01a13355-3ef4-7631-b5a0-c82930169f77@ 2026-10-13T11:37:55.732+08:00 (gpt-5.6-solmedium)01a13347-1ec4-7050-a430-57d78c0f289101a13347-6344-7d73-955c-6ab5e3ef92cd01a13348-01f0-7593-bf7f-a87175fef133
- Rate-limit snapshot beside failure (rollout):
plan_type=pro,used_percent=20, credits balance present,rate_limit_reached_type=null.
Please correlate uploaded thread 01a1337b-ddfb-7013-ba44-2716c02d2789 with account admission/routing.
Activity
github-actions commentedon Oct 13, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
86848696keny commentedon Oct 13, 2026
Business impact / severity (please escalate)
This is not an occasional inconvenience. It is blocking paid professional work during daytime hours.
I am a paying ChatGPT Pro subscriber on a high-capacity plan. Usage and credits are clearly available (~20% weekly used / credits remaining;
rate_limit_reached_type: null). Despite that, Local Codex on this account repeatedly returns:Selected model is at capacity/server_overloadedacross multiple models (
gpt-6-astraultra,gpt-5.6-solmedium, Fast off). The same machine and network works fine with a different ChatGPT account at the same time. Evenings often work; daytime is largely unusable. That pattern is incompatible with "just try another model / retry later" as the only guidance.How this is affecting work
server_overloaded— that is a costly dead end when the real issue is admission/routing, not quota.What I need
01a1337b-ddfb-7013-ba44-2716c02d2789).This has been going on across multiple days and is severely impacting my ability to work. Please prioritize.