Codex CLI on macOS: speech‑to‑text (Chinese + English+German) input not inserted into chat until pressing Enter (Cursor & Antigravity terminals)
What version of Codex is running?
codex-cli <your version here, e.g. 0.63.0>
What subscription do you have?
ChatGPT Pro
Which model were you using?
gpt-5.1-codex-max xhigh
What platform is your computer?
macOS
What issue are you seeing?
When I use speech-to-text on macOS to input a Chinese + English mixed sentence into Codex CLI, the recognized text does not appear in the Codex input line after the dictation finishes.
The text only appears (or gets sent as a full message) after I press Enter or trigger a newline / any keypress.
This seems to happen starting from the second conversation in the same Codex CLI session. The very first conversation usually behaves correctly.
The issue reproduces consistently both in:
- Cursor’s integrated terminal
- Antigravity’s Codex CLI integration
So it feels like a Codex CLI TUI / paste handling issue rather than an IDE-specific one.
What steps can reproduce the bug?
Rough repro steps:
- On macOS, start Codex CLI in a terminal (e.g. Cursor integrated terminal or Antigravity terminal) and sign in with my ChatGPT account.
- Start the first conversation:
- Place the cursor in the Codex input line.
- Use macOS speech-to-text / IME voice input to dictate a sentence that mixes Chinese and English.
- After dictation finishes, the text appears in the input line as expected. I can edit it and send it. (First conversation is OK.)
- In the same Codex CLI session, start a second conversation / follow-up:
- Again use the same speech-to-text workflow with Chinese + English mixed content.
- Actual behavior:
- After dictation finishes, the Codex input line remains empty, as if nothing was pasted.
- When I press Enter (or otherwise trigger a newline / keypress), the entire dictated text suddenly appears or is sent as a single message.
This pattern is reproducible every time for me, in both Cursor and Antigravity terminals.
What is the expected behavior?
- After speech-to-text finishes, the recognized text should be immediately inserted into the Codex input line, for every conversation (not only the first one).
- I should be able to see and edit the text before sending it.
- Pressing Enter should only submit the text, not “magically reveal” previously invisible input.
Additional information
- This behavior started only after the last two Codex / Codex CLI updates.
Before those updates, the exact same workflow (same Mac, same IDEs, same speech-to-text method) worked perfectly.
- I’ve already upgraded Codex CLI to the latest version and rebooted macOS; the issue still reproduces.
- Speech-to-text works fine in other applications (plain text editors, browser text inputs, etc.), so it doesn’t look like a system-level dictation problem.
I also noticed a very similar issue reported as “Codex CLI copy paste issue” (#7203) on the codex repo, where large pasted text does not appear until the user presses a key.
My case looks like the same underlying TUI / paste-handling problem, but on macOS with Chinese + English mixed speech-to-text input.
This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗