Can a decision model route requests between a cheap LLM and an expensive one?

Yes. TypeSafe’s own use-case map lists model routing as one of Jev’s built-in patterns: send it the prompt, have it classify intent and domain and estimate difficulty and risk, then let your code send the easy cases to a cheap model and escalate the rest to an expensive one. The router itself runs at Jev’s price, a small fraction of an LLM call, so the routing decision doesn’t eat into the savings it exists to create.

What comes back is a probability, not a verdict to trust blindly. The same threshold logic used for any Jev decision applies to a routing call too: a low-confidence pick belongs on the safer, costlier path rather than wherever Jev happened to guess. Running a cheap check ahead of every LLM call is the same math that favors a classifier over an LLM judge: pay a small fixed cost on every request so the expensive model only runs where the request actually needs it.