Not all models are equal

Every request you've sent so far named a model, and until now the choice didn't matter much. It does matter. The models available to you are not interchangeable: they differ in how fast they answer, how good the answer is on a hard problem, and how big the model is. Those three pull against each other.

The three things that trade off

• Speed. A smaller model returns its reply faster and can handle far more requests in the same amount of time. If you're calling the model in a loop over a thousand rows, speed is the whole game.
• Quality. On a genuinely hard task - tricky reasoning, careful code, a problem with several steps - a stronger model gets it right more often. On an easy task you often can't tell the models apart at all.
• Size. Bigger models (more parameters) tend to be more capable, but they cost more to run and answer more slowly. Size is roughly the price you pay for the extra capability.

You rarely get all three at once. A fast, small model and a slower, stronger model are two different tools, and picking well means knowing which one the job in front of you actually needs.

Why there isn't one "best" model

It's tempting to just always reach for the strongest model, but that's usually the wrong habit. On a simple task the strong model spends more time and (in production, where you pay per token) more money to produce an answer no better than the cheap model would have given you. And when you're firing off many requests, the slower model can turn a job that took seconds into one that takes minutes.

So the useful question is never "which model is best?" It's "which model is enough for this task?" The rest of this topic answers that: what each of the five lab models is good for, and how to match one to a task.

This is a different split from the one you met in the basics. There, the line was open versus frontier - who built the model and whether you can run it yourself. All five models here are open. The split now is about speed and strength among models you already have access to.

Your lab setup

This VM comes preconfigured with your AI keys - just type claude to get started. Your key and URL are already set as $OPENAI_BASE_URL and $OPENAI_API_KEY.