compare models

pick two or three: rates, context and capabilities side by side, and what the same workload costs on each.

or start from a comparison

one text model per lab, pairedusd / 1m tok · in / out