Repository navigation
Windows app hangs on splash screen and becomes Not Responding on Windows build 26200 #48148
Description
What version of the Codex App are you using (From “About Codex” dialog)?
26.917.9434.0
What subscription do you have?
ChatGPT Business
What platform is your computer?
Windows 11 x64, OS build 10.0.26200
What issue are you seeing?
What version of the Codex App are you using?
OpenAI.Codex 26.917.9434.0
Package:
OpenAI.Codex_26.917.9434.0_x64__2p2nqsd0c76g0
What platform is your computer?
Windows 11 x64
OS version reported by Codex log: 10.0.26200
What issue are you seeing?
The ChatGPT/Codex Windows desktop app consistently hangs during startup.
The window appears and reaches the OpenAI/ChatGPT splash screen with the logo in the centre, but then becomes unresponsive. Windows shows "ChatGPT (Not Responding)".
Task Manager shows multiple Codex/ChatGPT child processes still running. The main ChatGPT process typically consumes several hundred MB of RAM and remains stuck indefinitely.
The web version of ChatGPT works normally.
Relevant startup log
The application backend appears to start successfully.
The log reaches:
windows_core_runtime_launch_selected architecture=x64
Launching app
AppServerConnection Starting app-server connection
stdio_transport_spawned
window ready-to-show
window main frame finished load
initialize_handshake_result outcome=success
Codex CLI initialized
app_server_connection.state_changed ... next=connected
Host startup critical-path phases completed
The renderer also logs:
[electron-message-handler] [statsig-refresh-diagnostics] React root render requested
The app then remains frozen on the splash screen.
There is also this warning:
Failed to start the Artifact Session host:
"Artifact Session host Unix-socket transport is not available on Windows."
However, startup continues successfully after that warning.
There is also:
refresh_local_remote_control_client_id_failed
"Sign in to ChatGPT to check remote control authorization."
The app-server and Codex CLI had already successfully initialized before this.
Troubleshooting already performed
- Ended all ChatGPT/Codex processes in Task Manager
- Rebooted Windows multiple times
- Windows App Repair
- Windows App Reset
- Uninstalled and reinstalled ChatGPT/Codex
- Launched immediately after a clean reboot
- Tested launch with --disable-gpu
- Renamed ~/.codex/auth.json to force fresh authentication
- Renamed/cleared local Codex runtime/state locations
- Renamed LocalCache for the OpenAI.Codex package
- Temporarily disconnected non-essential USB/HID devices
- Closed SimHub, Simucube Tuner, iRacingManager and similar utilities
- Checked Windows Application event log for Application Hang/Application Error events
- Checked System event log for Display, DWM, nvlddmkm and WHEA errors
None of these changed the behaviour.
There were no Display/DWM/NVIDIA/WHEA errors around the failed launches.
Additional observations
The app consistently creates the window and initializes the local app-server successfully before the UI freezes.
This appears similar to other Windows 26200 reports where background processes start successfully but the desktop UI does not become usable.
The issue persists after reinstall/reset, suggesting it may be a renderer/UI regression rather than corrupted installation state.
What steps can reproduce the bug?
- Start Windows normally.
- Launch the ChatGPT/Codex desktop app.
- The application window opens and shows the OpenAI/ChatGPT splash screen.
- The splash screen remains visible and the UI never finishes loading.
- After a short time, Windows marks the application as "Not Responding".
- Task Manager shows multiple Codex/ChatGPT child processes still running while the main ChatGPT process remains hung.
The issue reproduces consistently after reboot, reinstall, reset, and with non-essential USB/HID devices disconnected.
What is the expected behavior?
The ChatGPT/Codex desktop app should complete startup after the splash screen and display the normal application UI, allowing me to sign in and use ChatGPT/Codex normally.
Additional information
The issue still reproduces after extensive troubleshooting.
Notably, the Codex backend appears to start successfully before the UI becomes unresponsive. Logs show the app-server connection succeeds, the main window reaches "ready-to-show", the main frame finishes loading, the Codex CLI initializes successfully, and the connection reaches "connected".
The failure appears to occur after the window/renderer starts, while the splash screen remains visible.
No related Display, DWM, NVIDIA, WHEA, or standard Application Hang errors were found in Windows Event Viewer during the failed launches.
The issue also persists after:
- Windows app Repair and Reset
- uninstall/reinstall
- multiple reboots
- --disable-gpu
- fresh auth state
- fresh Codex local/cache state
- disconnecting non-essential USB/HID devices
- closing SimHub, Simucube, iRacingManager, and similar utilities
The web version works normally.
Activity
github-actions commentedon Sep 25, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
Shogen95 commentedon Sep 26, 2026
Additional reproduction on Windows 11 x64 build 10.0.26200.
Codex App package initially reproduced on:
It then auto-updated during troubleshooting to:
The startup hang persists: the window remains on the loading/splash state and does not reach the usable UI.
Relevant behavior from the 26.924.1866.0 startup log:
This strongly matches the original report: package registration, local app-server, CLI initialization, and authentication all succeed, while the desktop UI fails to complete startup.
A later package check after the auto-update shows:
OpenAI.Codex 26.924.2738.0 Status=Ok
The web version continues to work normally.
Shogen95 commentedon Sep 26, 2026
Follow-up from the Windows 11 x64 build 10.0.26200 reproduction on Store package OpenAI.Codex 26.924.2738.0:
The CDP recovery workaround was attempted and did NOT restore the usable desktop UI on this machine.
The attempted path preserved MSIX package identity, launched with localhost-only remote debugging, waited for the app://-/index.html renderer, allowed startup to settle, and sent Page.reload with ignoreCache=true. The UI remained stuck afterwards.
This adds a negative result for the workaround on 26.924.2738.0: it can recover some affected machines (as reported elsewhere), but is not reliable across all Windows 26200 reproductions.
No app reset, reinstall, or deletion of local Codex data was performed during this attempt.
dariusbuinovskij commentedon Sep 26, 2026
I reproduced a similar startup freeze on Windows build 26200 with app version 26.924.2738.0.
Using Codex CLI and a local debugging connection, we identified a specific blocker:
Passing the actual reported backend version through the renderer’s existing initialization handler immediately changed readiness to "ready" and opened the main interface. This worked repeatedly. No application files, credentials, or persistent settings were manually edited.
The workaround is memory-only: restarting removed it and reproduced the freeze. This confirms the immediate blocking condition on my machine, but we haven’t determined why the initialization state was missing.
This may help investigate this issue. I’m including the Codex CLI instructions below; internal identifiers are build-specific and must be verified before applying the workaround.
Diagnose and temporarily unblock my ChatGPT Windows app, which stays on its startup spinner.
You may read local app logs and installed app code, launch/restart ChatGPT for diagnostics, and apply the memory-only workaround described below if you verify the same condition. Ask before editing files, changing persistent settings, reinstalling anything, or touching other applications.
lucasar-web commentedon Sep 26, 2026
Additional affected PC: 26.924.1866.0 / Windows 11 25H2 26200.9457
Adding an independent report, prepared with Codex diagnostic assistance and posted at the user's request. After the latest app update, the window stays indefinitely on the startup logo. The bundled CLI works when launched with
--no-daemon.Environment verified on 2026-09-26
OpenAI.Codex_26.924.1866.0_x64__2p2nqsd0c76g0; package status OK, not partially staged.26200.9457.0.158.0-alpha.2.32.0.31041.1004.ChatGPT.exefile version154.0.8037.57.Observed startup sequence (2026-09-26, UTC; selected log fields)
The approximately 19-minute launch contains no
[startup][renderer] app routes mountedentry. Read-only inspection confirmed that this marker still exists in the installed renderer code. This locates the incomplete startup stage but does not establish the underlying cause.Other checks
config.tomlparses successfully and contains zero NUL bytes. The global-state JSON and checked Chromium preference JSON files parse successfully.Scope and request
This is retrospective log analysis of user-observed failed launches. The GUI was no longer running at the process snapshot; no new GUI reproduction or live renderer-state inspection was performed. The null backend-version state and memory-only workaround described above have NOT been verified on this PC.
No reset, reinstall, cache clearing, authentication changes, process termination, or project/session-data edits were performed as part of this investigation. Conversation contents, tokens, account identifiers, private paths, raw dumps and full logs are intentionally excluded.
Please investigate the missing renderer initialization/ready transition and provide a supported recovery or fixed app version that preserves existing history. The shared Windows build is context, not evidence that Windows itself is the root cause.
raf-uwt commentedon Sep 26, 2026
Reposting the diagnostic from my preferred GitHub account.
Additional reproduction on Windows x64 / OS build 10.0.26200.
Current Store package during the latest tests:
Key finding: the startup hang is reproducibly tied to the presence of ~/.codex/config.toml, not to any specific TOML setting.
Failed-launch milestones on 26.924.2738.0:
The Artifact Session Unix-socket warning appears nonfatal because startup proceeds and the backend initializes successfully.
Safe workaround:
This suggests an existing-config/profile bootstrap path is preventing the renderer from completing startup, even when config.toml is empty. Please investigate the renderer initialization path for existing profiles/configs.