Provider arbitrage: paying frontier prices only for frontier questions
Model prices vary 50x for overlapping quality on easy queries. The arbitrage router exploits the spread automatically, with a quality bar you set per intent.
The model market has a spread any trader would recognize: frontier models cost an order of magnitude more than mid-tier ones, but on easy intents — factual lookups, simple classification, boilerplate transformation — their outputs are indistinguishable. Routing everything to the frontier means paying caviar prices for toast. Routing everything to mid-tier means failing your hard queries. The money is in routing well.
Crowkis Enterprise's arbitrage router does the routing with the cache's own intelligence: intent classification already grades query difficulty, so each query goes to the cheapest backend that clears your configured quality bar for that intent class. Easy traffic rides cheap models; genuinely hard reasoning rides the frontier; the bar, not vibes, decides.
The upgrade is a workflow, not a leap of faith.
The cache stacks multiplicatively underneath: repeated queries don't route anywhere — they hit. The router only prices the novel residue, and the cross-provider bridge means answers cached from one backend serve traffic on another, so the arbitrage never fragments your memory.
The bottom line
The combined effect on blended cost is dramatic and — unusually for cost work — quality-neutral by construction, since the bar is enforced per query. The spread is sitting there in every pricing page. Take it.