PRODUCTION SCENARIO
A team pairs a lower-cost executor with a frontier advisor to save money on chart-reading tasks. Shadow runs show the executor consults the advisor on nearly every task, and the bill now exceeds the frontier model alone.
Which approach gets the same quality for less?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: Drop the executor and run the advisor's model alone at medium effort
Consulting on nearly every task means paying advisor rates across the whole workload plus the executor's own tokens; the documented benchmark matched the advisor model alone at medium effort within run-to-run noise at about 2.6 times the cost, and the advice is to run that model itself. Caching helps little when every task carries a different chart. Dropping the advisor keeps the cheap executor and gives up the quality the pairing was bought for. Raising effort adds executor cost on top.