Choosing the Right Model for the Job

The best model is the cheapest one that clears your quality bar — and that changes per task.

Match the model to the work

• Hard reasoning, planning, ambiguous goals → a frontier model. This is where capability pays for itself.
• High-volume, well-defined steps (classify, extract, format) → a smaller, faster model. Often 80% of an agent's calls.
• Private or offline data → a local model, accepting a quality trade-off for control.

Mix models inside one agent

A common, money-saving pattern: a capable model plans and a fast model executes the routine sub-steps. You pay frontier prices only for frontier-shaped problems.

How to actually decide

Don't argue about it — run your golden eval set across two candidates and compare success, cost, and latency. The right answer is usually obvious within twenty examples.

Re-check quarterly

Model quality and pricing move fast. The choice you made six months ago deserves a fresh look — a routine that was too expensive last quarter may be trivial now.