Claude Settings: The 6 Dials That Change Every Answer

Claude is not average. Your settings are.

Six dials sit in plain sight inside the app, and most people touch one of them — the model picker — and only when something goes wrong. The other five sit at their defaults for months, quietly deciding how good every answer is.

None of them is hidden. None of them requires a plan upgrade to find. Here is what each one does, and, more importantly, when it is worth spending on.

The short version

• Effort = how long it thinks. Model = who is thinking.
• Voice and files = what it knows. Tools = what it can reach. Approvals = how far it goes alone.
• Before you send anything, ask one question: if this answer is wrong, what does it cost me?

1. The effort dial

Bottom right of the input box. Five levels: low, medium, high, extra, max.

It controls how hard Claude thinks before it answers. The crucial thing to understand is that it is a budget, not a default — effort costs time and it costs usage limits, so leaving it at max permanently is not "always getting the best answer," it is spending your allowance on tasks that did not need it.

A working rule:

• Low for shortening text, reformatting, and anything where you already know what the output should look like.
• Medium for first drafts — the bulk of ordinary work.
• Max for costly mistakes: the contract clause, the architecture decision, the thing you will build on top of for six months.

The dial is most valuable precisely where people forget it exists — on the expensive question asked in a hurry.

2. Model choice

Top half of that same menu: Haiku, Sonnet, Opus, Fable.

People conflate this with the effort dial, and they are genuinely different axes. Effort is how long it thinks. Model is who is thinking. A small model thinking hard and a large model thinking briefly are not interchangeable, and they fail in different ways.

Reach for Haiku when you already know the answer and just need it typed out, formatted or checked. Reach for Fable for your hardest calls — the ones where you are genuinely unsure and the cost of being wrong is real.

Everything in between is the daily middle, and the middle is where most work lives.

3. Voice input

The microphone. Overlooked because it looks like a convenience feature, and it is not.

Two things it changes. First, the obvious one: it moves about three times faster than your typing. Second, and far more important: it changes what you are willing to say.

When you type, you edit as you go. You compress. You leave out the background because writing it feels like effort. When you speak, you ramble — and messy, complete context gives Claude better clues than a tidy command does. The unstructured version of a problem usually contains the constraint that mattered, the one you would have trimmed out of a well-written prompt.

So use it deliberately: talk through the whole situation, badly, and let the model do the tidying.

4. Attach files

Upload the contract, the spreadsheet, the screenshot. Claude reads the actual file rather than your summary of it — which removes an entire category of error where the thing you left out of your description is the thing that mattered.

The upgrade most people miss: if you find yourself uploading the same file every week, move it into Projects instead. A file in a Project is available to every conversation in that Project without being re-attached, which turns a repeated chore into a standing piece of context.

That is the general pattern with this dial — attaching is for one-off documents, Projects are for the ones that define your work.