← All lessons
0270

Choosing a Model

Leaderboards crown a single best model, but real decisions have three axes: capability, cost, and speed — plus a fourth, privacy, that can override them all. A frontier API model wins on hard reasoning but costs the most per token and sends your data out. A small fast model is the right answer for routine work at high volume. A local open-weight model keeps everything on your hardware, capped by your GPU. The skill is matching the model to the task, not picking the biggest one.

Set what your task needs — difficulty, volume, privacy — and watch the recommended tier flip. Try: hard reasoning + low volume, then millions of routine calls, then anything with strict privacy.

Task difficultyroutinehard reasoning
Volumea few callsmillions/day
Privacy needpublic datamust stay local
Frontier model (API)BEST FIT

Best reasoning, highest price per token, data leaves your machine.

Small / fast model (API)

Cheap and quick; fine for routine tasks at scale.

Local open-weight model

Runs on your hardware; nothing leaves the machine; capability capped by your GPU.

There is no best model only a best fit. Move one slider and the answer changes.

Check yourself

Your task is summarising a million public support tickets a day. Why is the frontier model the wrong choice even though it is the most capable?

Go deeper (free): LMArena — community model leaderboard

Back to start →