[FEATURE] Customise `/model` selection

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

I sometimes want to use older versions of a model, due to their different performances at different effort levels. For example, Opus 4.6 on medium effort is _very_ useful for quick answers, where I would like a model slightly more intelligent than Sonnet, and shorter (and therefore faster) responses than Opus 4.7/4.8.

I've previously achieved this by using ANTHROPIC_CUSTOM_MODEL_OPTION and ANTHROPIC_CUSTOM_MODEL_OPTION_NAME. However, this only lets me access two versions of Opus (the default version & this custom model). As I have often found myself reaching for Opus 4.7 on high effort, I've chosen to instead switch my default Opus version to 4.7 using ANTHROPIC_DEFAULT_OPUS_MODEL, meaning that I need to manually type the Opus 4.8 model alias if I want to switch back to the latest model.

Proposed Solution

Add the ability to customise the models we can switch between using /model, preferably in either user or managed settings.

This could also eliminate the need to specify custom models via the environment variable, as custom models, names and descriptions could instead be placed in this new configuration for the /model list.

Even better would be a mechanism to specify the default effort for each of these versions when we switch to it using /model.

Alternative Solutions

• Enable multiple custom models to be defined using environment variables
• Other interface to allow switching the live version of a model in /model

Priority

Medium - Would be very helpful

Feature Category

Interactive mode (TUI)

Use Case Example

I'm working on a new complex feature:

1. I do some initial exploration with sonnet to build up context
2. I switch to Opus 4.6 on medium effort to compact, and ask some questions about the gathered context using /btw
3. I switch to Opus 4.8 on high effort to form a plan for this feature, or if I deem the feature to require less reasoning capability, I switch to Opus 4.7 (I've found that this can be cheaper per-prompt than 4.8)
4. I execute the plan (potentially with a clean context) using the best model for that task's complexity (potentially a cheaper / faster model)

Additional Context

_No response_