Picking the right model for the job
Model choice is a trade-off between quality, latency, price, context window, license and where data may travel. A frontier hosted model is often best for hard reasoning, while a smaller open-weight model such as Llama or Mistral can handle classification or extraction at a fraction of the cost and can run inside your own network. The shortlist also records license terms, since some open-weight models restrict commercial use or scale.
We shortlist two or three candidates, run them on your real inputs and compare scores, speed and cost per task. The answer is frequently a mix: a small model for routine steps and a larger one for the difficult minority. We document the reasoning so the decision can be revisited as models change. Distillation, where a large model teaches a smaller one, is another route to lower cost once quality is proven.
