Claude gives you a few dials. The good news: the defaults are well chosen, so you can ignore all of this and still do great work. But three of these dials are worth understanding, because reaching for the right one turns a slow, so-so answer into a fast, sharp one. They are the model, thinking, and web access, all set from the small picker next to the send button.
The models: fast, balanced, or heavyweight
Claude comes in a few sizes. They are not "good, better, best"; they are different tools for different jobs, trading speed against raw power. Think of a workshop: you do not reach for the sledgehammer to hang a picture.
Haiku is the sprinter. The fastest and lightest. Perfect for quick, simple jobs: a fast rewrite, a short summary, a straightforward question. When speed matters and the task is easy, Haiku is done before the others have cleared their throat.
Sonnet is the all-rounder, and the default. The best balance of speed and smarts, and the right choice for the large majority of your work: writing, summarising, everyday questions, light analysis. If you are not sure, this is the answer.
Opus is the heavyweight. Slower and more demanding of your usage, but it is what you want for genuinely hard work: knotty reasoning, deep analysis, serious coding. Bring it out when a task has real teeth.
Fable is the marathon runner. The most capable, built for the most demanding, long-running jobs. On most days you will not need it, but it is there when the work is big.
You can switch model in the middle of a chat, and the change applies to the next reply. So a fair habit is: stay on the default, and only reach up to the heavyweight when an answer comes back thin on a hard problem.
Thinking: when slower is smarter
Next to the model, the picker lets you turn on extended thinking. This tells Claude to reason step by step before it answers, working through the problem instead of replying off the cuff. It is the difference between a colleague who blurts the first thing and one who says "give me a second" and then nails it.
Turn it on for hard reasoning that needs no live facts: a multi-step logic puzzle, planning a tricky project, debugging code, weighing a thorny decision. Leave it off for the easy stuff, a quick email, a simple lookup, where thinking just adds a wait and spends more of your usage for no gain.
There is also an effort setting on the same menu, from low up to max. It is the same idea in a slider: higher effort means more thorough, slower, and heavier on usage. For everyday tasks the defaults are fine. Raise it only when a task clearly deserves the extra care.
Web access: for anything current
Claude's built-in knowledge stops at a cut-off date, so it does not know today's news, this week's prices, or a change made yesterday. For anything current, switch on web search (the + button in the message box, or just ask Claude to "search the web"). Claude looks things up live and hands back an answer with source links you can click.
One rule keeps you safe: trust, but verify. The sources are there for a reason. Claude can still misread a page or lean on a weak one, so on anything that matters, open a source and check it yourself. Let Claude do the finding; you own the fact. For a bigger dig across many sources, Claude also has a deeper Research mode, but plain web search covers most needs.
Matching the setting to the task
Put the three together and a simple habit falls out. Stay on the default for most things. When a problem is genuinely hard and self-contained, make Claude think. When you need something current, send it to the web and check the sources. That is the whole skill, and you can mix them: thinking and web search happily work together.
Do not fiddle for the sake of it. The single most common mistake is cranking everything to maximum on a simple question, which just makes you wait. Match the dial to the job.
Try it
Take one real question with a clear right answer, for example a bit of logic or a calculation from your work. Ask Claude once on the plain default, then ask the same thing again with extended thinking turned on. Read both.
You will usually see the thinking answer show its working and catch a nuance the quick one skated over. Now do the reverse test: ask something about this week (a price, a headline, a recent change) with web search off, then on. Off, Claude will hedge or rely on old knowledge; on, it fetches the facts with links. Two short tests, and the three dials stop being a mystery.
(General exercise. Role and industry versions come once accounts are in.)