---
id: models-thinking-and-web-access
lang: en
title: "Models, thinking, and web access"
seoTitle: "Claude models, thinking & web access"
description: "Claude's models and when to use each, extended thinking for hard problems, and web search for current facts, so you pick the right setting every time."
part: getting-started-with-claude
partTitle: "Getting started with Claude"
partOrder: 2
order: 3
level: Beginner
durationMinutes: 5
outcome: "Choose the right model for a task, switch on thinking when it helps, and use web access well."
summary: "Match the dials to the task: Sonnet for most work, Opus for the hardest, extended thinking for tough reasoning, and web search for anything current."
draft: false
updated: "2026-07-26"
furtherReading:
- title: "When should I use web search, extended thinking, and research?"
url: "https://support.claude.com/en/articles/11095361-when-should-i-use-web-search-extended-thinking-and-research"
note: "Anthropic's own guide to picking the right one."
- title: "Change the model, effort, and thinking settings"
url: "https://support.claude.com/en/articles/8664678-change-the-model-effort-and-thinking-settings"
note: "Exactly where the picker lives and what each setting does."
- title: "Enable and use web search"
url: "https://support.claude.com/en/articles/10684626-enable-and-use-web-search"
note: "Turning web search on and reading the sources."
- title: "Claude models overview"
url: "https://platform.claude.com/docs/en/about-claude/models/overview"
note: "The current line-up and what each model is built for."
keywords:
- Claude models
- Claude Opus vs Sonnet
- Claude Haiku
- extended thinking
- Claude web search
- which Claude model to use
quiz:
intro: "Five quick questions. Ace them all and the next lesson unlocks."
passScore: 5
questions:
- q: "For most everyday work, drafting, summarising, answering questions, which model should you reach for?"
options:
- "Always the biggest one, whatever it is called"
- "Sonnet, the balanced default"
- "Haiku, always"
- "It does not matter at all"
answer: 1
explain: "Sonnet is the all-rounder and the default for good reason: the best balance of speed and smarts for the bulk of daily tasks. Save the heavyweight for genuinely hard jobs."
- q: "You have a knotty, multi-step problem to reason through, and it needs no up-to-date information. What is the best setting?"
options:
- "Turn on web search"
- "Turn on extended thinking so Claude reasons before it answers"
- "Use the fastest model with no extras"
- "Ask the same thing five times"
answer: 1
explain: "Extended thinking lets Claude break the problem down and weigh options before replying. It shines on hard reasoning, tricky analysis, and planning, where no live facts are needed."
- q: "You need today's figures and recent news for a market summary. What do you switch on?"
options:
- "Extended thinking"
- "Web search"
- "Nothing; Claude always knows the latest"
- "The biggest model, which makes it current"
answer: 1
explain: "Claude's built-in knowledge has a cut-off date, so anything newer than that has to come from the live web. Web search fetches it and hands back sources you can click."
- q: "Claude searches the web and gives you an answer with source links. On anything that matters, what should you do?"
options:
- "Trust it completely; the links prove it is right"
- "Click the sources and check the claim yourself"
- "Ignore the links"
- "Assume it is wrong and start over"
answer: 1
explain: "The links are there so you can verify. Claude can still misread a page or lean on a weak source, so on anything important, open the source and check. Let it do the finding; you own the fact."
- q: "You have a quick, simple task and want to be light on your usage. What is the sensible choice?"
options:
- "The heaviest model on maximum effort"
- "The fastest model (Haiku) or a lower effort setting"
- "Extended thinking, to be safe"
- "Run it on every model and compare"
answer: 1
explain: "Simple jobs do not need deep reasoning. The fast model, or a lower effort level, is quicker and stretches your usage further. Match the effort to the task."
pass:
badge: "Perfect score"
title: "You can pick the right Claude now."
body: "Full marks. You know which model to reach for, when to make Claude think, and when to send it to the web. Next up: artifacts, where Claude builds work you can keep and reuse."
retry: "Check the marked answers and give it another go. You've got this."
---
> **Level:** Beginner ยท **Reading time:** ~5 min
>
> **What you'll be able to do:** pick the right model for a task, switch on thinking when it helps, and use web search well.
>
> *Last checked: July 2026. Model names and numbers change often; the ideas here (a fast tier, a balanced default, a heavyweight) are the durable part. Check the [current line-up](https://platform.claude.com/docs/en/about-claude/models/overview) for today's exact names.*
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.)*
## Key terms
[Model](/glossary#model), [Extended thinking](/glossary#extended-thinking), [Web search](/glossary#web-search), [Token](/glossary#token)