Repository navigation
Account-level "Selected model is at capacity" on 100% of requests across all models (Pro plan) #46185
Description
Environment
- OS: Windows 11 (10.0.26200)
- Codex CLI 0.153.4; the desktop app behaves identically
- Plan: Pro (
chatgpt_plan_type=pro), account idd8927685-…(full id available privately on request) - All tests: 2026-09-17, 07:54–08:05 UTC, one machine, one network
What happens
Every request, on every model, is rejected with:
Selected model is at capacity. Please try a different model.
Model-by-model probe (one minimal request each, --sandbox read-only, low reasoning effort)
| model | result |
|---|---|
| gpt-6-astra | at capacity |
| gpt-5.6-sol | at capacity |
| gpt-5.6-luna | at capacity |
| gpt-reserve | at capacity |
| gpt-5.5 — 3 consecutive attempts | at capacity (3/3) |
gpt-5.5 with service_tier=priority |
at capacity |
| gpt-5.6-terra | stream disconnected before completion, request id 8977beed-af35-4524-8d94-61c25248e952 |
Reproduce: printf 'say OK' | codex exec --skip-git-repo-check -m gpt-5.5 -c model_reasoning_effort=low --sandbox read-only
Ruled out
- Client: CLI and desktop app fail identically
- Client version: the rejection is server-side, not client-side
- Model choice: every model is rejected, including gpt-reserve
- Machine and network: unchanged; no usage-limit message is ever shown, only the capacity rejection
Ask
This looks like account-level admission/routing rather than model-level capacity. Please check this account's admission state, and whether activity-based throttling has been applied to it. The account also runs automated agent workloads, so if admission is limited by account activity, please point me at the applicable limits.
Independent reports describe the same account-level pattern — e.g. #46103: on one machine and one network, one account fails 0/11 while another account succeeds 11/11.
Activity
github-actions commentedon Sep 17, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
lixiang4543 commentedon Sep 17, 2026
I suspect that they have planted spies in the GPT6 model.The reason is that after switching to the GPT6 model, this error occurred, causing all models to become unavailable. Before this switch, everything was working fine.
cdevsystems commentedon Sep 17, 2026
Error running remote compact task: Codex ran out of room in the model's context window. Start a new thread or clear earlier history before retrying.
starRiverKoi commentedon Sep 17, 2026
One additional detail that may be important: this has not been a short-lived or single-session issue.
The affected Plus account has been experiencing the same
Selected model is at capacity/server_overloadedbehavior for approximately three consecutive days.
During that period I have tried:
codex execinvocations--ephemeral--ignore-user-config
The behavior continues to follow this Plus account.
Meanwhile, the Free account on the same PC and network can successfully use the same
gpt-5.6-lunamodel.
So if the cause is region/data-center routing, it may be worth checking whether the affected account has some kind of persistent/sticky affinity to a degraded region or serving pool, rather than this being a one-time routing decision.
The persistence across roughly three days is the main reason I suspect this may be more than a transient regional capacity event.