When reviewing a diff an commenting, there ought to be a model selector.

When reviewing a diff an commenting, there ought to be a model selector.

Often I find that, when I am submitting feedback on some code, it would be handy to switch to Auto model selection, or to a cheaper model, and need to un-maximize, switch or check what model is in use, then switch back. Many times I will also need to switch to a "smarter" model too, which runs into the same issue.

Even worse, if something is already running in the session, then I can't reliably switch models for just that one piece of feedback.

The constant back and forth this causes, and the waiting it imposes on me as a reviewer, mean that I can either stick with one smart model, stick with one cheap model, or suffer from constant delays while I wait for the agent to complete it's current task. None of those options are ideal.

Having the ability to choose which model will handle a given message, especially on the diff viewer, would solve this by allowing to instruct the use of a specific model for a given message.

The overall session ought to have a "default" model for handling messages that arrive without a specified override, as well as handling messages from other sessions.

Start in the /chat diff-review flow and trace the message composer and existing model-selection controls. Clarify how a per-message override should coexist with the session default, including messages sent while another task is running. Done means reviewers can choose a model for an individual message without leaving the diff view, while unqualified messages use the session default.