Not every call needs your most powerful model. This lesson is about matching the model to the job across three axes: capability, cost, and latency. Reasoning models are stronger and slower and dearer; fast models are cheaper and quicker and weaker.

The skill is knowing which each task needs, and you practise it live by swapping models on the same endpoint and watching the tradeoffs move - same prompt, different model, and seeing quality, cost, and speed shift. Most production systems use a mix: the strong model where it matters, the cheap one everywhere else.