You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
What version of the Codex App are you using (From “About Codex” dialog)?
Version 26.616.71553 • Released Jun 22, 2026
What subscription do you have?
Pro $100/mo
What platform is your computer?
Windows 11
What issue are you seeing?
When trying to send a message as simple as "Hello" I receive a pop up message stating "Unable to send message" "Loading..." and the only option is to click OK.
What steps can reproduce the bug?
Click "New Chate"
Type "Hello" in message box
Pop-up error occurs.
Changing models does not fix issue.
What is the expected behavior?
Click "New Chat"
Type "Hello" in message box
Codex ChatGPT should respond to message.
Additional information
I have tried repairing the Codex Windows Desktop app, I have uninstalled and reinstalled the app, I have updated the app, problem continues. This is happening on 2 of my Windows 11 computers.
Can confirm on macOS latest Sequioa, on Codex Version 26.623.31921 (Released Jun 26, 2026). This is bananas, our company cannot work... ALL models and all speed settings have the same effect.
EDIT: I think the bug is related to Codex own app upgrades and ongoing longer chats, meaning that Codex upgrades basically break the possibility to continue a chat. Which is insane for a software that has nearly daily auto-updates, built by a 850 billion dollar company.
Also seeing this on macOS. However in my case, this is not a permanent send failure for the thread.
Codex Desktop: 26.623.31443, build 4441.
Auth mode: API key login.
Behavior:
Open an existing thread.
Send the first message.
Modal appears: Unable to send message / Loading....
That initial message is not sent.
After dismissing the modal, the same thread starts allowing normal sends.
So it looks like the first send attempt is failing/dropping the message, but it also initializes enough thread/session state that subsequent sends work.
Local logs near the failure show Codex turn/start / turn/steer routes returning errorCode=null, while renderer fetches to ChatGPT account endpoints like /backend-api/wham/usage and /backend-api/wham/tasks/list return 401 Unauthorized with hadToken=false.
Now the updates are blocking any Intel Mac from even using the Codex app. This is insane! 850 billion dollar company. we are paying $100 per dev. Now totally locked out
Cancel the forking to new worktree after 2 - 3 seconds
Retry sending message in the stale conversation
Conversation may work again (it did for me, hope for your too).
Small offtopic note on this (quote of user @panique):
This is insane! 850 billion dollar company. we are paying $100 per dev. Now totally locked out
We feel the same. As a cybersecurity company, we think the product and models are strong, and Altman’s intention to make AI publicly accessible seems noble. But customer/product management- and especially support, is really poor, which is a farce for a company of this size.
For example, one of our employee’s identity verifications failed because their name contained a special character. We contacted support, but were sent in circles and, in the end, they failed to resolve the issue and even got insolent. The account is still unusable, no solution has been offered to this day, and the ticket was simply closed.
We will be moving our company to Anthropic in Q4 this year. Also, Anthropic’s stance against the use of AI in warfare and weaponry seems, to me personally, like a healthier intent for humanity’s future with AI.
FIXED! The issue seems to be fixed in latest Codex Version 26.707.51957, released 12th of July 2026.
Also the Codex app has been rebranded into the "ChatGPT app", so it's now not a standalone app anymore.
I recommend @thefin0614 to retest and potentially close the github issue as fixed!
Activity
github-actions commentedon Jun 23, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
panique commentedon Jun 26, 2026
Can confirm on macOS latest Sequioa, on Codex Version 26.623.31921 (Released Jun 26, 2026). This is bananas, our company cannot work... ALL models and all speed settings have the same effect.
EDIT: I think the bug is related to Codex own app upgrades and ongoing longer chats, meaning that Codex upgrades basically break the possibility to continue a chat. Which is insane for a software that has nearly daily auto-updates, built by a 850 billion dollar company.
Audacity88 commentedon Jun 26, 2026
Also seeing this on macOS. However in my case, this is not a permanent send failure for the thread.
Codex Desktop: 26.623.31443, build 4441.
Auth mode: API key login.
Behavior:
Unable to send message/Loading....So it looks like the first send attempt is failing/dropping the message, but it also initializes enough thread/session state that subsequent sends work.
Local logs near the failure show Codex
turn/start/turn/steerroutes returningerrorCode=null, while renderer fetches to ChatGPT account endpoints like/backend-api/wham/usageand/backend-api/wham/tasks/listreturn401 UnauthorizedwithhadToken=false.thefin0614 commentedon Jun 26, 2026
CodexIssue.md
Attached is a Claude Code session I used to fix this issue. Please review
panique commentedon Jun 27, 2026
Now the updates are blocking any Intel Mac from even using the Codex app. This is insane! 850 billion dollar company. we are paying $100 per dev. Now totally locked out
Haxel0rd commentedon Jul 7, 2026
Quick fix that worked for me:
Conversation may work again (it did for me, hope for your too).
Small offtopic note on this (quote of user @panique):
We feel the same. As a cybersecurity company, we think the product and models are strong, and Altman’s intention to make AI publicly accessible seems noble. But customer/product management- and especially support, is really poor, which is a farce for a company of this size.
For example, one of our employee’s identity verifications failed because their name contained a special character. We contacted support, but were sent in circles and, in the end, they failed to resolve the issue and even got insolent. The account is still unusable, no solution has been offered to this day, and the ticket was simply closed.
We will be moving our company to Anthropic in Q4 this year. Also, Anthropic’s stance against the use of AI in warfare and weaponry seems, to me personally, like a healthier intent for humanity’s future with AI.
alsherif-khalaf commentedon Jul 10, 2026
same issue In Mac
alsherif-khalaf commentedon Jul 10, 2026
I found the issue , just delete .codex/config.toml
panique commentedon Jul 12, 2026
FIXED! The issue seems to be fixed in latest Codex Version 26.707.51957, released 12th of July 2026.
Also the Codex app has been rebranded into the "ChatGPT app", so it's now not a standalone app anymore.
I recommend @thefin0614 to retest and potentially close the github issue as fixed!
Thanks everybody ;)
etraut-openai commentedon Aug 4, 2026
Send failure fixed in Desktop 26.707.51957.