Codex CLI recommends gpt-5.4, then fails with 'Selected model is at capacity' despite available quota
已打开 02:21PM - 07 Apr 26 UTC
已关闭 04:37PM - 07 Apr 26 UTC
bug
# Codex CLI GPT-5.4 Capacity Incident Report
Date: 2026-04-07
Reporter context:… ChatGPT-authenticated Codex CLI user
## Summary
The user was running Codex CLI with `gpt-5.4` after Codex surfaced an upgrade path from
`gpt-5.3-codex` to `gpt-5.4`. Despite the user's report that 5-hour and weekly Codex
quota remained largely available, a turn failed with:
`Selected model is at capacity. Please try a different model.`
This appears to be a model-capacity/admission failure rather than quota exhaustion.
## Local evidence
- CLI version: `codex-cli 0.118.0`
- Auth mode: `Logged in using ChatGPT`
- Project config model: `gpt-5.4`
- Global Codex host config model: `gpt-5.4`
- Upgrade notice persisted locally:
- `~/.codex-host/config.toml` contains:
- `[notice.model_migrations]`
- `"gpt-5.3-codex" = "gpt-5.4"`
- `~/.codex-host/models_cache.json` contains upgrade text:
- `Introducing GPT-5.4`
- `You can always keep using GPT-5.3-Codex if you prefer.`
- Capacity error log:
- `~/.codex-host/log/codex-tui.log`
- Timestamp: `2026-04-07T13:41:04.538491Z`
- Message: `Turn error: Selected model is at capacity. Please try a different model.`
## Reproduction notes
- A later minimal request using the same ChatGPT-authenticated Codex CLI and the same
`gpt-5.4` model succeeded.
- A minimal request using `gpt-5.3-codex` also succeeded.
- This suggests a transient `gpt-5.4` capacity/admission problem, not a persistent
account-wide quota exhaustion.
## Why this is problematic
- The product encouraged migration from `gpt-5.3-codex` to `gpt-5.4`.
- The user is on a paid tier and reported substantial remaining quota.
- The failure mode did not explain whether this was:
- a quota problem,
- an abuse/security restriction,
- a temporary model-capacity shortage, or
- a workspace/account entitlement mismatch.
- Public documentation does not disclose the model-capacity priority order for Pro,
Business, Enterprise, or Edu users in Codex CLI.
## Requested clarifications from OpenAI
1. For this incident, was the failure caused by:
- temporary model capacity,
- an entitlement routing issue,
- an abuse guardrail,
- account security review, or
- another internal policy?
2. Which entitlement path was actually applied to this local Codex CLI session?
- personal Pro
- Business workspace
- Enterprise/Edu workspace
- some other ChatGPT-linked path
3. If this was capacity-related, what priority policy applied?
4. Why was a user with available quota shown a model-capacity failure instead of a more
explicit explanation?
5. If `gpt-5.4` is promoted as the preferred upgrade from `gpt-5.3-codex`, what
reliability expectation should users have in Codex CLI?
## Public documentation reviewed
- Using Codex with your ChatGPT plan
- https://help.openai.com/en/articles/11369540-using-codex-with-your-chatgpt-plan
- GPT-5.3 and GPT-5.4 in ChatGPT
- https://help.openai.com/en/articles/11909943-gpt-53-and-54-in-chatgpt
- Troubleshooting Model Feature Access Issues
- https://help.openai.com/en/articles/10258669-troubleshooting-model-feature-access-issues
## Constraints encountered while attempting submission
- The official support path exposed in the Help Center is the `Chat now` widget or
phone support.
- From this terminal environment, the authenticated Help Center chat widget could not be
programmatically opened and submitted end-to-end.
白芸汐 03-25 20:551楼
sunfly 03-25 20:573楼
loyicnaup 03-25 20:584楼
良芳 无 03-25 21:368楼
liusx 03-25 21:399楼
Curing 03-25 21:4110楼
vme50kksl 03-25 21:4111楼
问就是很帅 03-25 21:4412楼
Delilah Moore 03-25 21:4714楼
mebrian 03-26 14:3817楼
gopython3 04-22 15:3318楼
去去毒 04-22 15:5319楼