app-server supports set model_provider on `turn/start`

[caibirdme]

What variant of Codex are you using?

0.122.0

What feature would you like to see?

in thread/start, thread/resume, they both support {model, modelProvider}, while in turn/start, it just supports {model}.

Additional information

No response

[github-actions]

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

• #Switching models mid-session does not update provider/endpoint — Ollama endpoint persists after switching to gpt-5.4 #18764
• #Model switch mid-conversation fails: "encrypted content could not be decrypted" #17541

Powered by Codex Action

[etraut-openai]

The core harness doesn't generally support switching model providers in the middle of a thread. We recommend starting a new thread if you switch providers.

Can you explain the use case that's prompting this feature request?

[caibirdme]

The core harness doesn't generally support switching model providers in the middle of a thread. We recommend starting a new thread if you switch providers.

Can you explain the use case that's prompting this feature request?

We subscribed gpt-5.4 directly from openai platform, and subscribed other gpt-x models from azure(we have discount). It's normal for people to switch model among 5.4-high,5.4-mini,5.3-codex back and forth. But the problem is, we have two providers(openai & azure)