Bug: "Model not supported when using X-OpenAI-Internal-Codex-Responses-Lite" error with latest models #30461

What version of the Codex App are you using (From “About Codex” dialog)?

26.623.42026

What subscription do you have?

plus

What platform is your computer?

Mac OS

What issue are you seeing?

After the recent update to the Codex desktop client, I am constantly hitting an error when trying to use the latest models (e.g., GPT-5.5 series) or specific configurations. The error message states:
This model is not supported when using X-OpenAI-Internal-Codex-Responses-Lite.

What steps can reproduce the bug?

1. Open the Codex Desktop application.
2. Select the latest model from the model picker (e.g., GPT-5.5 or the default configuration that triggers the issue).
3. Type any simple prompt in the chat input box (e.g., "Hello, write a simple function").
4. Press Enter to send the prompt.
5. The request fails immediately with the error message: "This model is not supported when using X-OpenAI-Internal-Codex-Responses-Lite."

Note on requested metrics:

• Session ID: (If applicable, you can find this in your client logs or settings. Leave blank or N/A if not visible).
• Token limit usage / Context window usage: N/A (The error triggers immediately upon sending the initial prompt, before any tokens are processed or generated by the model).

What is the expected behavior?

The Codex Desktop client should handle the API request seamlessly without throwing a transport protocol error.

Specifically:

1. If the selected model does not support the X-OpenAI-Internal-Codex-Responses-Lite transport header, the client should automatically fallback to the full responses path.
2. Alternatively, the client should strip the incompatible X-OpenAI-Internal-Codex-Responses-Lite header when making requests to these specific model endpoints.
3. The prompt should be sent successfully, and the model should return the text/code response as expected.

Additional information

• This issue seems to have started occurring after a recent update to the client.
• According to the built-in AI support assistant, the root cause is that the client enforces the Lite Codex Responses transport layer, which is strictly incompatible with the newest model endpoints. The support assistant recommended either switching to a supported model or manually removing the X-OpenAI-Internal-Codex-Responses-Lite header.
• As a temporary workaround, switching the model picker back to older versions (like gpt-4o or gpt-5.4) temporarily resolves the issue, but the error consistently blocks the use of the latest model configurations.