Windows app launches stale cached app-server runtime (0.142.0) instead of current one — causes "model requires a newer version of Codex" and corrupts shared models_cache.json

Open 💬 2 comments Opened Jul 11, 2026 by PROJECT-OCULUS

💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

What version of the Codex App are you using (From "About Codex" dialog)?

26.707.3748.0 (Microsoft Store, package OpenAI.Codex)

What subscription do you have?

Pro

What platform is your computer?

Microsoft Windows NT 10.0.26220.0 x64

What issue are you seeing?

The app rejected gpt-5.6-sol in an existing conversation with:

The 'gpt-5.6-sol' model requires a newer version of Codex. Please upgrade to the latest app or CLI and try again.

No update was available: winget upgrade --id 9PLM9XGG6VKS --source msstore and the Store's own update scan (WinRT AppInstallManager.SearchForAllUpdatesAsync) both reported the app current. The same model worked fine from the standalone CLI 0.144.1 on the same machine and account (codex exec -m gpt-5.6-sol → OK).

Root cause I traced on my machine:

ChatGPT.exe spawned its backend as

%LOCALAPPDATA%\OpenAI\Codex\bin\38dff8711e296435\codex.exe -c features.code_mode_host=true app-server --analytics-default-enabled
That binary embeds CLI version 0.142.0 (provisioned Jun 26).

A current runtime was sitting right next to it — %LOCALAPPDATA%\OpenAI\Codex\bin\a7c12ebff69fb123\ embeds 0.144.0 (provisioned Jul 9) — and the app package itself bundles 0.144.0 in app\resources\codex.exe. The app still launched the stale hash. This looks like the same stale-versioned-path pattern reported in #30270, but for the core app-server runtime rather than plugins.
Because the server gates models by client version, the 0.142.0 app-server was denied gpt-5.6-sol, producing the misleading "upgrade" error on an up-to-date install.
Cross-client side effect: the stale app-server periodically rewrote the shared ~/.codex/models_cache.json stamped "client_version": "0.142.0" and without the gpt-5.6 model entries — degrading model metadata for every other Codex client on the machine that shares that cache. I suspect this mechanism is behind the "Model metadata for gpt-5.6-sol not found; defaulting to fallback metadata" reports in #31826 and #31869.

What steps can reproduce the bug?

I can't force the app to pin a stale runtime on demand, but the broken state is easy to detect when present:

# 1. Which binary is the app's backend actually running?
Get-CimInstance Win32_Process -Filter "Name='codex.exe'" |
  Where-Object { $_.CommandLine -like '*app-server*' } |
  Select-Object ProcessId, ExecutablePath
# parent process is ChatGPT.exe

# 2. Embedded CLI version of each cached runtime (no version resource on the exe;
#    grep the binary for its version string)
Get-ChildItem "$env:LOCALAPPDATA\OpenAI\Codex\bin" -Directory | ForEach-Object {
  $exe = Join-Path $_.FullName 'codex.exe'
  if (Test-Path $exe) {
    $v = Select-String -Path $exe -Pattern 'version: [0-9.]+' -AllMatches -Encoding oem |
         ForEach-Object { $_.Matches.Value } | Select-Object -First 1
    "$($_.Name)  $v"
  }
}

# 3. Who last wrote the shared models cache, and which models it contains
Get-Content "$env:USERPROFILE\.codex\models_cache.json" -Raw |
  ConvertFrom-Json | Select-Object client_version, fetched_at,
  @{n='slugs';e={$_.models.slug -join ', '}}

Broken state observed: app-server running from a hash dir embedding 0.142.0 while a 0.144.0 hash dir exists; models_cache.json stamped client_version: 0.142.0 with no gpt-5.6-* slugs.

What is the expected behavior?

The app should launch the app-server from the newest provisioned runtime (or the binary bundled in its own package), and never let an older cached runtime rewrite the shared ~/.codex/models_cache.json with a reduced model list.

Additional information

Workaround that fully resolved it for me:

Quit the app.
Rename the stale runtime dir and the poisoned cache:

%LOCALAPPDATA%\OpenAI\Codex\bin\38dff8711e296435 → *.bak, ~/.codex/models_cache.json → *.bak

Relaunch. The app-server then ran from the packaged 0.144.0 binary (C:\Program Files\WindowsApps\OpenAI.Codex_26.707.3748.0_...\app\resources\codex.exe), the cache regenerated with client_version: 0.144.0 and all gpt-5.6-* slugs, and gpt-5.6-sol worked again in the app — including in the previously failing conversation.

Related: #31826 (same error on latest version, no cause identified), #31869 (same error on up-to-date Linux CLI; macOS fine — consistent with a poisoned local cache/state rather than the binary), #30270 (stale versioned-path pattern for plugins), #20821 (older Windows-app variant with gpt-5.5).

View original on GitHub ↗

2 Comments

github-actions[bot] contributor · 1 month ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

#31835
#32449
#32146
#31869
#31826

Powered by Codex Action

croc81 · 26 days ago

GPT-5.6 models disappeared from the Codex coding-agent picker after a Windows app update

I may be experiencing the same Windows Codex app regression described in this issue.

Environment

ChatGPT Windows app version: 26.727.51351
App release date shown in the About dialog: August 1, 2026
About dialog: “Powered by Codex & OWL”
Subscription: ChatGPT Plus
Operating system: Windows 11, x64, OS build 26200
Project type: Local Codex projects
Repository location: WSL Ubuntu repositories accessed directly through paths such as \\wsl.localhost\Ubuntu-22.04\...
Authentication: ChatGPT sign-in
Models cache client version: 0.142.5

Issue

GPT-5.6 Sol was explicitly available in the actual coding-agent model picker and successfully used in my Local Codex project named trade-alert-sol approximately two days before this issue started.

I used GPT-5.6 Sol for a real coding task in that project. It analyzed the repository and updated project documentation successfully.

After installing the Windows application update released on August 1, 2026, all GPT-5.6 models disappeared from the coding-agent model picker in every Codex project.

The affected projects include:

trade-alert-sol, where GPT-5.6 Sol had previously worked
light-cms-sol, another Local project pointing directly to a repository inside WSL Ubuntu

The project coding-agent selector now displays Custom Light. Opening the selector only shows:

GPT-5.5
GPT-5.4
GPT-5.4 Mini

GPT-5.6 Sol, GPT-5.6 Terra and GPT-5.6 Luna are all missing.

GPT-5.6 Sol is still selectable in the separate Quick Chat window inside the same Codex application. However, it is missing from the actual coding-agent model picker in the project composer.

Quick Chat and the project coding agent therefore appear to receive different model-availability lists.

More importantly, GPT-5.6 Sol did not merely appear in Quick Chat. Before the update, it was explicitly selectable and successfully used by the actual coding agent in my trade-alert-sol project.

Steps to reproduce

Sign in to the Windows application with the affected ChatGPT Plus account.
Open an existing Local Codex project pointing directly to a repository inside WSL Ubuntu.
Open the coding-agent model picker at the bottom of the project composer.
Observe that the selector displays Custom Light.
Open the selector.
Observe that it only offers GPT-5.5, GPT-5.4 and GPT-5.4 Mini.
Open the separate Quick Chat window in the same application.
Observe that GPT-5.6 Sol is available in Quick Chat but not in the project coding-agent picker.

The problem affects multiple projects, including an existing project where GPT-5.6 Sol previously worked and a newly configured Local project.

Expected behavior

GPT-5.6 Sol should remain available in the Codex coding-agent picker because it had already been enabled and successfully used with the same ChatGPT Plus account and installation.

The coding-agent model picker and model cache should not revert to an older or reduced model list after an application update.

If the model configured in config.toml is unavailable, the application should display a clear error or availability message instead of only showing Custom Light.

Diagnostic evidence

My user-level configuration file at C:\Users\jasu\.codex\config.toml explicitly contains:

model = "gpt-5.6-sol"
model_reasoning_effort = "low"

The repository-specific .codex directories do not contain a config.toml overriding the selected model.

I inspected the user-level model cache using this read-only PowerShell command:

Get-Content "$env:USERPROFILE\.codex\models_cache.json" -Raw | ConvertFrom-Json | Select-Object client_version, fetched_at, @{Name='slugs';Expression={$_.models.slug -join ', '}}

The result was:

client_version: 0.142.5
fetched_at: 2026-08-01T14:13:02.090877600Z
models: codex-auto-review, gpt-5.5, gpt-5.4, gpt-5.4-mini

The recently fetched cache does not contain:

gpt-5.6-sol
gpt-5.6-terra
gpt-5.6-luna

This confirms that the problem is not limited to visual filtering in the model picker. The model list stored in models_cache.json itself does not contain the GPT-5.6 models.

Troubleshooting already completed

Logged out of the application and signed back in.
Completely quit the application, including the system-tray process.
Confirmed through Windows Task Manager that no ChatGPT, OpenAI or Codex processes remained running.
Restarted the application and signed in again.
Confirmed that the problem affects multiple Codex projects.
Confirmed that the projects point to the correct WSL repositories.
Confirmed that the repositories are trusted.
Confirmed that no repository-specific config.toml overrides the selected model.
Confirmed that the user-level config.toml explicitly selects gpt-5.6-sol.
Ran Settings > Configuration > Diagnose, which reported: Codex dependencies look healthy.
Ran Settings > Configuration > Reinstall, which reported: Codex dependencies were reinstalled.
Checked the coding-agent model picker again after reinstalling the dependencies. GPT-5.6 was still missing.
Inspected the current models_cache.json, which still contains only the older model list.

Additional information

The issue began immediately after installing Windows application version 26.727.51351, released on August 1, 2026.

GPT-5.6 Sol had been available and working in the coding-agent picker approximately two days earlier. Before that session, I had only seen GPT-5.5 in Codex. GPT-5.6 Sol then became available, worked successfully in a real Codex coding session, and disappeared again after the update.

The fact that this affects multiple projects makes a repository-specific configuration problem unlikely.

The fact that GPT-5.6 Sol remains available in Quick Chat indicates that the same ChatGPT Plus account can access the model through another interface inside the application. However, the strongest evidence is that GPT-5.6 Sol had also already been available and functional in the actual project coding agent before the update.

I submitted the application logs separately through /feedback.

Feedback ID:

019fb329-6a9a-7bf0-adfb-eeac6deb597f

This Feedback ID may help correlate this GitHub report with the submitted application logs.

The symptoms appear related to this issue and to the stale or reduced model-cache behavior described in issue #33146.

I have not independently confirmed whether the Windows application is launching an older cached codex.exe runtime. Therefore, I cannot confirm that the underlying cause is identical, but the observed model-cache behavior and symptoms are very similar.