Choosing a model

People say “using Claude” as if it were one thing. It isn’t. Claude is a family of models that share the same training and the same values but differ a great deal in how deeply they think, how fast they answer, and how much they cost to run. Picking between them is one of the quietest high-leverage habits available to you, and almost nobody does it — which is why most people are simultaneously overspending on trivial work and under-resourced on the hard stuff.

The rule that covers ninety per cent of cases is: match the brain to the difficulty of the task. Sonnet is the default and deserves to be — it handles most drafting, explaining, analysing, and building perfectly well. Haiku is the sprinter for jobs where you already know what you want and just need it done. Opus and Fable are for the problems with real depth, the ones where you’ve tried twice and kept missing. The guide and capability page below lay out the four tiers side by side with the specific signals for moving up or down.

Switching is not a commitment. It happens mid-conversation — the thread continues, only the model behind it changes — so the practical move is to notice a signal and act on it rather than deciding once at the start. Stuck after two attempts? Go up a tier. Churning through a dozen small edits? Drop to Haiku. Genuinely unsure? Stay on Sonnet; the default is well chosen.

Two things sit next to this choice and are easy to confuse with it. How hard Claude thinks is a separate dial from which model is doing the thinking — see thinking and effort, which you can turn up without changing tier. And the reason model choice shows up on your bill at all is covered in what it costs; a bigger model on a small task is the single most common source of quiet waste.