Models Aren't One Thing

Model families

Part of: Choosing the Right Model

You don't drive a moving truck to pick up a coffee. But when people first reach for an AI model, they grab the biggest, newest, most expensive one for every job, including jobs a tiny model would handle in a fraction of the time and cost. What it is There is no single "the AI." Every provider ships a family of models, and within that family several tiers, usually something like small, medium, and large. They share a lineage but differ in size, speed, price, and how hard a problem they can handle. A useful mental split: - Small / fast models : cheap, quick, good at simple, well-defined work. - Large / smart models : pricier, slower, but stronger on hard reasoning and long, messy tasks. The key idea: a tier is just a labeled point on a spectrum. You aren't choosing "good vs bad." You're choosing which trade-offs you want for this task. How it works Picture the tiers as a lineup you can pick from at request time: Each tier is the same kind of thing, text in and text out, so your code barely changes between them. You swap a label, and the cost/speed/smarts profile changes underneath. Why it matters Treating "the AI" as one fixed object leads to expensive, slow apps: - Newer isn't autom

Challenge: Tier Selector