GPT-5.4 swapped English words for Arabic

Resolved 💬 2 comments Opened Mar 20, 2026 by mstallone Closed Mar 21, 2026
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

What version of Codex CLI is running?

0.116.0

What subscription do you have?

Pro

Which model were you using?

gpt-5.4

What platform is your computer?

Darwin 24.6.0 arm64 arm

What terminal emulator and version are you using (if applicable)?

Ghostty

What issue are you seeing?

In the output, the model generated an Arabic word in an English sentence while working in an English repository

On worker restart, startup reconciliation, or alarm reconciliation, a launch that is still داخل the intended recovery window can be downgraded to failed early

I originally thought this was a tokenization/unicode rendering problem; however, the standalone translation of that word makes sense in the context of the sentence (though, I could be wrong as I don't speak Arabic).

What steps can reproduce the bug?

Uploaded thread: 019d0d24-9bd7-7770-abd3-20f7e54b9b6d

What is the expected behavior?

If working with Codex in a specific language, the model should not change languages.

Additional information

_No response_

View original on GitHub ↗

2 Comments

github-actions[bot] contributor · 5 months ago

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

  • #13823
  • #13867

Powered by Codex Action

etraut-openai contributor · 5 months ago

Thanks for the bug report. We'll share the feedback with the team responsible for training our codex models. We're tracking this feedback and similar reports so we can improve future models. And thanks for using the /feedback command. That additional information is useful.