[Windows ARM64] ChatGPT desktop 26.707.9981 (arm64) crashes on every launch — ChatGPT.exe exports no N-API symbols, delay-load of napi_create_function fails (0xC06D007F)

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

• #33415
• #33392
• #33381
• #33385
• #33384

Powered by Codex Action

Confirming the same root cause on another Windows 11 ARM64 machine (24 GB RAM), from an independent minidump analysis — with one extra data point: the fatal N-API call doesn't have to come from the serialport addon. In my case it comes from windows-updater.node, and the trigger looks server-side.

• Build: OpenAI.Codex_26.707.9981.0_arm64__2p2nqsd0c76g0, installed 2026-07-14 from the Store.
• The same build ran fine in hours-long sessions on Jul 14–15. Starting the night of 2026-07-16 every launch dies ~10 s in — 6/6 launches (05:06–05:25 local), all ending at the same bootstrap phase. No reboot since Jul 9, no local changes (system Node.js arm64 v24.14.1 untouched since March).
• Crashpad dump (browser process, ptype=browser): exception 0xC06D007F in KERNELBASE, DelayLoadInfo.szDll = "node.exe", missing proc = napi_create_function — identical to this issue.
• ChatGPT.exe (arm64) contains no napi_create_function string anywhere in the binary, so it cannot be in the export name table — matches your PE export walk.
• In my dump the only native addons loaded were the app's own: better_sqlite3.node and windows-updater.node. No serialport/device-kit module was loaded yet. windows-updater.node delay-imports node.exe and references napi_create_function.
• Main desktop log right before death (UTC): 01:25:15.247Z Codex runtime config selected artifact=codex-primary-runtime … → 01:25:16.403Z warning [install-primary-runtime] primary_runtime_update_poll_failed errorMessage="Failed to download primary runtime manifest (404 Not Found)." → last entry 01:25:17.073Z, Crashpad dump written the same second.

So the missing N-API exports in the arm64 host exe are the root cause (as this issue states), and any code path that loads an N-API addon kills the browser process — updater bootstrap in my case, serialport in yours. The "worked for two days, then crash-loops overnight with zero local changes" pattern supports the server-side rollout trigger suggested in #33384.

Environment exclusions from my side: no third-party modules in-process (all 115 modules in the dump are Microsoft/OpenAI — rules out AV interference), no memory-pressure events that day, no OS servicing between last-good and first-bad launch.

Happy to share the .dmp privately if useful.

I'm facing the same issue.

Device name	Surface Pro 12
Processor	Snapdragon X Plus (8-core) @ 3.30 GHz (3.24 GHz)
Installed RAM	16.0 GB (15.6 GB usable)
Graphics card	Qualcomm(R) Adreno(TM) X1-45 GPU (128 MB)
System type	64-bit operating system, ARM-based processor
OS        Windows 11 Home 25H2

Confirming on Snapdragon X2 / Windows 11 (build 28000), Codex desktop 26.707.9981.0 (arm64) — identical 0xC06D007F delay-load failure with node.napi.node on the stack, crashes ~2s into every launch (Crashpad dump each time).

Triage datapoint: this is not an OS-side/update issue. I rolled back the July 15 Windows cumulative update (28000.2525 → 28000.2269) and rebooted — still crashes identically. Reinstalling the same build also reproduces it. Began ~15h after the app self-updated to 26.707.9981.0.

Confirmed on another Windows 11 ARM64 system (Apple Silicon VM).

• Package: OpenAI.Codex_26.707.9981.0_arm64
• Fresh Store reinstall does not resolve the issue.
• Crashpad generates a dump on every launch.
• Application log shows successful renderer initialization before abruptly stopping.

My dump also references:

@serialport/bindings-cpp/prebuilds/win32-arm64/node.napi.node
In my case the dump also reports a missing export related to addon registration (node_register_module_v143), which appears consistent with the missing N-API export diagnosis described here.

Same here Surface Laptop 7 ARM64

我也是，surface pro11 xplus ARM64，这几天的变化就是前几天我的电脑系统自动更新了，以及codex自动更新了，然后今天发生的情况就是打开后会出现几秒界面，然后就闪退，没有任何报错

Confirmed on OpenAI.Codex 26.707.12708.0 ARM64.

Environment:

• Windows ARM VM on Apple Silicon Mac
• Process: ChatGPT.exe

WinDbg results:

• Exception: 0xC06D007F / ERROR_PROC_NOT_FOUND
• Faulting module: node.napi.node
• Module:

@serialport/bindings-cpp/prebuilds/win32-arm64/node.napi.node

• Delay-load target: node.exe
• Missing export: napi_create_function
• Failure bucket:

APPLICATION_FAULT_c06d007f_node.napi.node!Unknown

Reinstalling, resetting the app, and starting with a fresh .codex profile
did not resolve the crash. Crash occurs approximately 6 seconds after launch.

Seeing the same issue on:

Processor Snapdragon X Plus (8-core) @ 3.30 GHz (3.24 GHz)
Installed RAM 16.0 GB (15.6 GB usable)
Graphics card Qualcomm(R) Adreno(TM) X1-45 GPU (128 MB)
System type 64-bit operating system, ARM-based processor

ChatGPT Version: 26.707.12708.0

Was working fine yesterday until i updated from Codex to ChatGPT

I am seeing a closely related failure on Windows x64, although the symptom is different from the ARM64 startup crash.

Environment:

Codex Desktop: OpenAI.Codex_26.707.12708.0_x64__2p2nqsd0c76g0
Windows: Microsoft Windows NT 10.0.26100.0 x64

Symptoms:
Opening the Codex sidebar repeatedly causes an approximately 10-second UI stall, high disk activity, and repeated Windows Error Reporting events. Codex recovers afterward, but the issue reproduces consistently.

WER / dump analysis:

ExceptionCode: 0xC06D007F

Delay-load failure:
DLL: node.exe
Missing procedure: napi_module_register
Last error: 0x7F (ERROR_PROC_NOT_FOUND)

The failing native addon appears to be:

@worklouder/device-kit-oai
-> @worklouder/wl-device-kit
-> serialport
-> @serialport/bindings-cpp
-> serialport.node

I do not own or use Work Louder hardware.

This appears related to #33429, but on Windows x64 with a different missing N-API symbol (napi_module_register instead of napi_create_function).

The bundled Work Louder / Codex Micro device integration seems to be initialized automatically and can trigger repeated crash/retry behavior.

I can provide additional WER records or crash metadata if needed.

Here's a temporary workaround that addresses the problem:
https://gist.github.com/triesmon/bdc20a18ec34108e2608c1294c451f86

Note this can also cause extreme lag when sending messages and switching chats on normal machines. This isn't just an ARM64 issue. On x86_64 it spams the Windows event log and makes the UI extremely laggy. I can open a separate issue if need be. I've 100% isolated the issue to work louder.

Possible workaround (worked for me on ARM64)

I wanted to share what finally resolved this for me in case it helps someone else.

Environment

Surface Pro 11 (Snapdragon X Elite)
Windows 11 Insider Beta 26220.8764 (ARM64)
Microsoft Store version of ChatGPT
Existing Codex installation migrated to ChatGPT

Symptoms

Codex worked previously.
After the migration to ChatGPT, the app would either:
display the "What's New"/onboarding screen for a few seconds and exit, or
exit immediately after launch.
Browser ChatGPT worked normally.

During troubleshooting I also noticed Microsoft Store crashing, although I'm not certain whether that was related.

What finally fixed it

A normal uninstall/reinstall was not enough.

The important difference was completely removing all remaining user data.

I manually deleted every remaining OpenAI/Codex folder I could find, including:

%USERPROFILE%\.codex - RENAME - OLD.codex
%USERPROFILE%\.cache - RENAME - OLD.cache
%USERPROFILE%\.agents - RENAME - OLD.agents
%LOCALAPPDATA%\OpenAI - RENAME - OLD.OpenAI
Any remaining OpenAI/Codex folders under %LOCALAPPDATA%
Any remaining OpenAI package data I could locate

Only after removing all of the above did I reinstall ChatGPT.

The key observation

Every previous reinstall preserved my login and immediately returned to the existing state.

After deleting all remaining folders, ChatGPT behaved like a completely fresh installation:

prompted me to sign in again
displayed the full first-run onboarding
requested Windows integration permissions
completed setup successfully
now launches normally

This makes me suspect that some migration/profile state survives uninstall and can prevent the new ChatGPT desktop app from starting correctly.

I can't say with certainty which folder contained the problematic state, but removing all remaining OpenAI/Codex user data before reinstalling resolved the issue for me.

I can confirm that @MarkMasiak solution worked for me too. Thank you!

I can confirm that @MarkMasiak solution worked for me too. Thank you!

I've had Codex scan the Old Folders and it has restored the chats and work too. i got it to recreate everything in to new replacement projects.

Codex/ChatGPT created a 'Recovered project and chat map.MD' file - so i could decide what to do.

Update: resolved by build 26.707.91948 — no user-data wipe required

The Microsoft Store delivered ChatGPT desktop 26.707.91948 (arm64) on 2026-07-16, and the app now launches normally on the same Windows-on-ARM machine. Signed-in state, projects, and threads all survived intact.

For anyone finding this issue with the same symptoms (instant exit on launch, no logs, Crashpad dump showing 0xC06D007F / napi_create_function / node.exe):

• The crash was specific to build 26.707.9981 (arm64), whose ChatGPT.exe shipped without N-API exports.
• You do not need to uninstall or delete any user data — just update the app in the Microsoft Store (Updates & downloads → Check for updates). Re the workaround in the comment above: a full data wipe + reinstall most likely worked because the reinstall pulled the fixed build, not because of the removed folders.

Leaving it to the maintainers whether to close this or keep it open to track how the arm64 build lost its N-API exports in CI.

Really appreciate all the feedback! We have addressed the issue https://github.com/openai/codex/issues/33375#issuecomment-4991172145.

是的，这个问题直接在微软商店更新后解决了 2026/07/17