Which model for which job: don't send a jumbo to fetch coffee

You wouldn't hire a structural engineer to hang a picture. Picking an AI model is the same call: match the muscle to the job.

Most tools let you pick a model. Fast ones are plenty for small edits; slow, thinking ones earn their cost on the hard problems.

Models come, roughly, in three grades: fast-and-cheap, capable-all-rounder, and reasoning models that think step by step before answering (that's the reasoning-model lesson). Bigger isn't automatically better; it's better for harder work, and slower and pricier for easy work.

The reason it matters is cost and speed, which are the same thing you metered in the hosting lesson. Reasoning "thinks" in tokens you pay for; Anthropic's extended-thinking docs describe higher thinking budgets as buying "more comprehensive reasoning, with diminishing returns… at the cost of increased latency." A jumbo model on a rename is pure wait and spend.

We pick this way building this very site: a fast model for small edits and copy, a heavier one for tricky logic, and a reasoning mode for the stubborn bugs and anything touching security. The skill isn't always using the biggest model; it's knowing when you don't need it.

Try it with the rules off. Think of your last three AI requests. Which actually needed deep thinking, and which were renames or copy tweaks a fast model would nail instantly? Most people find they were overpaying for the easy ones.