I built a cli model router on top of @typesafeai called Frost. You pass it any prompt, short or long, and it recommends a model, harness, and effort level. It is extremely cheap and fast, around 200 to 250 ms end to end after one TypeSafe request.
It can pull the latest model data from online sources like Artificial Analysis, or you can configure it to your own preferences, and optional CodexBar hooks let recommendations respect Claude, Codex, and other subscription limits you have wired up.