Free sample lesson ยท Which one should I be using for this?

Not A Ladder

The most capable model is the wrong default for most of what you do.

Lesson 1 of the Wondering course Which one should I be using for this?

The default that costs you

Most people pick a tool once, pick the best model in it, and never think about it again.

That is a reasonable default and it is wrong in both directions. You wait for a frontier model on tasks a fast one handles instantly, and you use a fast model on the analysis that actually needed care.

The fix is not learning every model. It is having three or four questions you ask before you start, which take about ten seconds.

The three tiers

Providers offer roughly the same shape.

Fast. Cheap, quick, capable of most routine language work. Classification, extraction, simple rewrites, formatting, first-pass summarising.

Balanced. The sensible default for real work. Drafting, analysis, most of what you do.

Frontier. The most capable, slowest, most expensive. Genuinely better on hard reasoning, long complex documents, and unusual problems.

The names change constantly. The shape does not, which is why learning the shape beats memorising the names.

Where the fast model wins

It is worth being concrete, because people under-use this tier badly.

The fast model is the right choice when the task is simple but repeated. Four hundred items to classify. Every email to be tagged. A long document chunked and processed piece by piece.

On that kind of work the frontier model is not more accurate in a way you can measure, and it is several times the cost and several times the wait.

Use the expensive model where the difficulty is, not everywhere.

The thinking question

Separately from tier: should it think for longer before answering?

Yes when the problem has multiple steps and a checkable answer. Working through constraints. Debugging. Analysis where one wrong step invalidates the rest. Planning with dependencies.

No when the task is fluent production. Drafting, rewriting, tone, summarising, brainstorming. You would not have shown your working either, and the extra time buys nothing.

A useful test: could you check whether the answer is right? If yes, thinking helps. If it is a matter of taste, it does not.

Keep going

This is lesson 1 of 3 in Which one should I be using for this?. Next up:

  1. The Surface Matters More
  2. Benchmarks Are Not Your Task

Start this course free

Wondering turns any topic into short lessons you can read, watch, or listen to, with practice questions and an AI tutor.