Teams running 35+ models in production is wild but it tracks.
They live across a team's whole portfolio, not just inside one app. Coding agents want the best reasoning, support bots want fast & cheap, research tools need long context.
Different jobs pick different winners.