Model routing is the thrift of the moment. The idea: do not send every request to the expensive model. Route the easy work to a cheap one, save the costly tokens for the hard problems. It is sensible. It is also the single oldest instinct in any child who has ever been handed a fistful of arcade tokens.
You do not spend the good token on skee-ball. You spend the brass one.
The arcade always knew
Every token holder has practiced model routing without naming it. The common brass token goes into the machine that returns the most tickets per play. The rare token, the gold one, the prototype, is routed nowhere. It is held back, because some tokens are too valuable to spend on a low-stakes query.
Behold a token you would never route into a coin slot.
Tiers, but permanent
The AI version of this idea has a flaw the arcade version does not. When you route a request to the cheap model, the cheap tokens are still consumed and still vanish. You have merely chosen a less expensive way to own nothing.
Route a token here and the tiering survives the transaction. This catalog holds 846 tokens across every tier of value, from the common workhorse to the rarities that top out at $2,398.92. The cheap ones are still here. The dear ones are still here. Routing, in a sane economy, does not destroy the thing routed.
The verdict
Spend the common token freely and guard the rare one. To learn which tokens belong in the guarded tier, read about control tokens and rare compositions, or return to the field guide to tokens.
In brass we trust.