Codex "Model not found" on Windows: Chat breaks, Work keeps working
Back to the blog
Articletroubleshootingcodexopenai

Codex "Model not found" on Windows: Chat breaks, Work keeps working

MafraNovember 21, 20265 min read

You open Codex on Windows, pick a model, send any message, and get Model not found. That message covers two different problems, and what decides which one you have is not the error text: it is where it appears. One of them is fixed in 30 seconds. The other cannot be fixed on your machine.

This guide separates the two cases, shows the command for each, and tells you what is not worth trying.

What does the Codex "Model not found" actually mean?

The request reached the backend with a model identifier that path does not recognize. It is not a network issue, not a blocked account, and not a usage limit. What changes everything is which part of the product fails, because Codex has three surfaces that pick a model through different paths: Chat in the Windows app, Work, and the CLI.

Where the error appearsWhat keeps workingCauseWhat to do
Chat in the Windows appWork and Codexthe desktop Chat model catalogtest another time zone
CLI in default modethe same CLI with --no-daemondaemon out of sync with the CLIrun without the daemon
Every path, alwaysnothingmodel retired or account without accessswitch models

Chat fails while Work keeps working: what is going on?

This is the most common case on Windows and the most confusing one, because the product looks half broken. It is not: Chat and Work use separate model catalogs in the desktop app, and only the Chat one is affected.

The same pattern shows up across three open issues in the public Codex repository in October 2026: #54083 shows both model versions, 5.6 and 6, failing at all three reasoning levels while the web responds; #54231 records that uninstalling, reinstalling, signing out and resetting the app change nothing; and #55574 confirms that switching models inside Chat does not help either.

So if your symptom is a broken Chat with a working Work, stop trying to reinstall. It is not the installation.

The trigger you can actually reproduce: time zone

The most useful clue is in #54340, from October 15, 2026. The error appears consistently when Windows is set to the China Standard Time zone, and disappears when you switch to another zone with the same UTC+8 offset. Four zones were tested:

Windows time zoneOffsetChat
China Standard TimeUTC+8fails
Singapore Standard TimeUTC+8works
Taipei Standard TimeUTC+8works
W. Australia Standard TimeUTC+8works

That is bizarre, and that is exactly why it is worth testing before anything else: if the local time does not change but the error disappears, you confirmed the diagnosis in a minute.

To switch it from PowerShell, without going through the UI:

# see the current time zone
Get-TimeZone

# switch to a UTC+8 zone that works
Set-TimeZone -Id "Singapore Standard Time"

# confirm the change
Get-TimeZone

Close the Codex app completely and open it again before testing. Reopening just the conversation does not reload the model catalog.

The CLI returns model_not_found and --no-daemon works: why?

In the CLI the error looks different. Instead of Model not found, you get a 400 with model_not_found in the body:

{"type":"error","status":400,"error":{"type":"invalid_request_error",
 "message":"The 'gpt-6.1-sol' model is not supported when using Codex with a ChatGPT account.",
 "code":"model_not_found"}}

Issue #59646, opened on November 5, 2026, documents the pattern: the CLI is on version 0.166.0, but the background daemon still runs app-server 0.158.0, and that is what serves the request. The old version does not know the new model, so it answers that the model does not exist.

The test that confirms it is running the same CLI without the daemon:

codex --no-daemon

If the model answers that way, the problem is the stale daemon, not your login and not your account. Worth knowing: deleting the %USERPROFILE%\.codex\app-server-daemon folder and restarting did not fix it in that issue report. The daemon came back up on the old version.

How do you confirm which of the two cases you have in a minute?

Run the commands in order and read the output before changing any setting.

# 1. which CLI version is installed
codex --version

# 2. what the local diagnostic reports
codex doctor --summary

# 3. test the same model without the daemon
codex --no-daemon

If step 3 works and default mode does not, it is the daemon. If the error only appears in the app Chat and the CLI answers normally, it is the Chat catalog, and the path is the time zone. If all three fail, the model was retired or the account lost access to it.

What does not fix this error

  • Reinstalling the app. Issue #54231 tried a full uninstall and the error stayed.
  • Switching models in the Chat picker. #55574 shows every combination fails.
  • Signing out and back in. Also tested in #54231, no effect.
  • Deleting the whole config folder. You lose profiles and providers, and in the daemon case it comes back on the same version.
  • Waiting for the next update as the only way out. --no-daemon fixes the CLI today, without waiting for a release.

How to remove this class of error from your path

Both cases share a root: an intermediate layer decides which model serves your request, and when that layer is stale or out of sync you lose the session. In the CLI daemon it is even visible to the naked eye as a version mismatch. The way not to stop working because of it is having a second path ready, instead of depending on a single layer to keep going.

If your day stops because one provider's model layer broke, the next step is having somewhere to continue. Meet Verboo Code.

Sources: open issues in the public Codex repository, #54083, #54231, #54340, #55574 and #59646, checked on November 21, 2026.

Enjoyed this article?
Share knowledge with your network.
// Read also

Related articles