How to use Thinking on Codex provider

[Paololalap]

E.g I set the thinking to medium but when I'm choosing a model on CLI tools it's not indicating what i selected as thinking type or is it automatically reflected when i set to medium and choose the model on CLI tools?

[yhay81]

Yes — with the current implementation it is applied automatically. The CLI's model picker only shows the model identity; it is not intended to echo 9router's provider-level thinking override.

The request path works like this:

1. When the Codex provider is configured as medium, 9router injects reasoning_effort: "medium" into each routed request if the client did not already send its own reasoning_effort. See chatCore.js.
2. The Codex executor converts that to the Responses API shape:

   {
     "reasoning": {
       "effort": "medium",
       "summary": "auto"
     }
   }

   It also requests reasoning.encrypted_content. See open-sse/executors/codex.js.
3. The plain model name is sent upstream after any thinking suffix is removed, so seeing only e.g. gpt-5.6-sol in the CLI selector is normal.

The precedence is important:

client reasoning.effort
> client reasoning_effort
> model-name effort suffix
> executor default

The provider-level setting is injected before that chain, but only when the client did not specify reasoning_effort. Therefore a CLI that explicitly sends its own effort can override the dashboard setting.

To verify it rather than relying on the model picker, make one request and inspect the 9router request log/details. The correlated request line includes THINK:medium; the forwarded request should contain reasoning.effort = medium. If it does not, confirm that the selected CLI model is actually routed through the Codex provider and inspect whether the CLI supplied a different effort.

[Smartraining96]

Anybody...

why model in gemini reduced as pictured, is that normal? previuosly all model are available.