Native multi-provider model switching in Codex Desktop

What variant of Codex are you using?

Desktop App

What feature would you like to see?

Codex Desktop currently treats model_provider as a thread/config-level setting, which makes switching between models from different providers inside the same conversation unreliable.

I built a working local workaround using a single logical provider (router) backed by a localhost routing service:

deepseek-* models → DeepSeek Responses API
GPT models → ChatGPT/Codex backend
Codex always sees the same model_provider = "router"

Because Codex already supports changing the model mid-thread, this allows me to switch between GPT and DeepSeek from the normal model picker inside the same conversation while preserving the full context.

The model picker now effectively looks like:

GPT-6 Astra
DeepSeek Flash
DeepSeek V4 Pro
GPT-5.6 Sol
GPT-5.6 Terra
GPT-5.6 Luna
GPT-5.5

This suggests Codex could support this natively by making a picker entry represent something like:

{
model,
model_provider,
authentication
}

rather than requiring one provider to remain active for the whole thread.

Expected UX:

Start a conversation using DeepSeek.
Switch to GPT from the model picker.
Continue the same conversation and context.
Switch back whenever desired.

No duplicate threads, separate CODEX_HOMEs, or manual config changes.

I tested the router approach successfully: DeepSeek requests reached api.deepseek.com and GPT requests reached the ChatGPT Codex backend while remaining in the same Codex thread.

It would be great to have first-class multi-provider routing/model switching directly in Codex Desktop.

Additional information

No response

Start by tracing how model_provider and the normal model picker are represented for Desktop App threads; the issue does not name files or tests. Compare the router workaround with the requested {model, model_provider, authentication} picker entry, and verify completion by switching DeepSeek and GPT in one conversation while preserving context and routing each request to its provider.