Initial new-chat conversation ignores selected model and falls back to GPT-5.5
Resolved 💬 3 comments Opened May 3, 2026 by gaduviera Closed May 5, 2026
💡 Likely answer: A maintainer (github-actions[bot], contributor)
responded on this thread — see the highlighted reply below.
Summary
In a brand-new chat, the first conversation appears to ignore the selected model and defaults to GPT-5.5 (medium intelligence), even when the user tries to switch before sending the first message.
Observed behavior
- This only happens in the initial / first conversation state of a new chat.
- After the first message is sent, the model selector appears to behave normally.
- The UI may show a different selection, but the first turn still executes with GPT-5.5.
Expected behavior
The model selected by the user before the first message should be the model used for that first turn.
Reproduction steps
- Open a new chat.
- Before sending the first message, open the model selector.
- Choose a model other than GPT-5.5.
- Send the first message.
- Observe that the first turn still uses GPT-5.5 / medium intelligence.
- Send a follow-up message and observe that the selector then appears to work normally.
Environment
- Variant: Codex App (desktop app)
- Platform: Windows
Notes
This looks like a startup / initialization precedence issue in the first conversation state rather than a general selector bug.
3 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
Which codex variant are you using? The app? CLI? extension? web?
Which version and platform?
The test was performed using the Codex App (desktop app), not the CLI, extension, or web version.
Platform: Windows.
Issue actualizado