Preferences Over Benchmarks: Model Routing — Archana Kamath & Tyler Gillam, DigitalOcean
Benchmark-chasing is the wrong model selection instinct; task-specific routing beats one-model-fits-all
“There is no single best model. The right one depends on the actual request.”
DigitalOcean's VP of inference engineering argues that routing requests to the right-sized model per task — not chasing benchmark leaderboards — is the correct mental model for production AI. She cites three forcing functions: exploding inference costs (even Walmart, Uber, and Microsoft are capping usage), poor fit of frontier models for simple tasks, and reliability risk from single-model dependency. The framing that AI cost optimization is arriving in months rather than the 15 years it took for cloud is a useful signal about how fast this discipline is maturing.