Windows 11 Insider Build 26200: ChatGPT Desktop continuously flickers before login, renderer remains alive (OpenAI.Codex 26.715.7063.0)

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

• #34299
• #34219
• #34141

Powered by Codex Action

i have the exact same issue . win 11 rtx 5090 latest version OpenAi.Codex , latest nvidia studio driver

I can also reproduce this on ChatGPT Desktop 26.715.52143 (Microsoft Store).

The issue still exists after updating from 26.715.31925 to 26.715.52143.

Likely root cause found: deactivated_workspace account-bootstrap retry loop (not GPU)

Hit identical symptoms on stable Windows 11 25H2 (26200.8875, Intel UHD Graphics): main window flickers continuously / flashes black, sidebar stuck on skeleton placeholders, Settings > log out does nothing. Web ChatGPT and ChatGPT Classic worked fine on the same machine. Affected app versions here: 26.715.7063.0, 26.730.8199.0 and 26.803.5235.0 (updating did not help).

None of the usual suspects changed anything: --disable-gpu, --disable-direct-composition, Windows transparency effects off, reinstall, Store update. Since the original reporter has an RTX 4070 and this machine is Intel iGPU-only, the GPU was never the common factor.

The desktop logs (%LOCALAPPDATA%\Packages\OpenAI.Codex_2p2nqsd0c76g0\LocalCache\Local\Codex\Logs) show what is actually happening:

warning [electron-message-handler] sa_server_request_failed ... errorMessage={"detail":{"code":"deactivated_workspace"}}
info    [electron-message-handler] [chatgpt-account-lookup] completed accountsHttpStatus=402 authenticatedAccountPresent=false
error   [electron-message-handler] Statsig: error while bootstrapping post-login client ...
warning [electron-message-handler] chatgpt_pubsub_connection_failed ... deactivated_workspace

This block repeats 15-20 times per second for the whole session (11,801 hits in one 13-minute log; every flickering session shows thousands). The signed-in account's Business workspace had lapsed billing, and the backend blanket-402s every request in that state (same server-side behavior as #16292). The app retries the account bootstrap forever and every retry repaints the UI, which renders as the endless flicker. Because every backend call fails, the logout control is dead too, so users are stuck.

Workaround that fully fixed it:

1. Quit ChatGPT (system tray icon > Quit).
2. Back up and remove the cached auth/session state: rename %USERPROFILE%\.codex\auth.json and %LOCALAPPDATA%\Packages\OpenAI.Codex_2p2nqsd0c76g0\LocalCache\Roaming\Codex\web (rename, don't delete, so it's reversible).
3. Relaunch, sign in again, and pick an account/workspace that is active.

Result: zero deactivated_workspace hits in a 2-hour session afterwards, no flicker, app fully usable.

Suggestion for the fix: when account bootstrap receives deactivated_workspace / HTTP 402, break the retry loop and show an actionable error or workspace switcher instead of repainting indefinitely. This may also explain #34219 and #34299 for users whose Team/Business workspace lapsed while a personal subscription stayed active.

1. Quit ChatGPT (system tray icon > Quit). 2. Back up and remove the cached auth/session state: rename %USERPROFILE%\.codex\auth.json and %LOCALAPPDATA%\Packages\OpenAI.Codex_2p2nqsd0c76g0\LocalCache\Roaming\Codex\web (rename, don't delete, so it's reversible). 3. Relaunch, sign in again, and pick an account/workspace that is active.

Thanks, that worked.

Renaming the cached auth/session files fixed the blinking issue.

Hopefully this can serve as a temporary workaround for others too.