The complete Claude guide
Every plan, model and product. Verified 23 September 2026.
Hey, it's Adam. Claude is the AI I use most in my own business, and it's also the hardest one to get a straight answer about, because it changes every few weeks. This page covers every plan, model and product that's live right now, what each costs, and what Claude still can't do. No affiliate links and no email form.
Checked against Anthropic's own documentation on 23 September 2026. That's the day after Anthropic released Claude Opus 5.5, so any guide dated before 22 September is already a model behind. Where Anthropic doesn't publish a number, I say so instead of guessing. Where Anthropic's own pages contradict each other, I show both versions.
Start here: Claude is a family of products now 🧭
If you think of Claude as a chat box, you're seeing about a fifth of it. Here's what's under the name today:
- Claude: the main app on web, desktop and mobile. Chat and Cowork are being merged into it (more on that below). This is where you ask, write, research, and hand off longer jobs that end in real files.
- Claude Code: the coding agent. It works in your terminal, your editor, the desktop app and the web, with access to your real files.
- Claude Design, Claude Slides and Claude Docs: three separate products for visual work, presentations and documents. All three are in beta and on paid plans only.
- Claude in Chrome: a browser extension that can read pages, click, type and fill forms for you.
- Claude for Microsoft 365: add-ins inside Excel, PowerPoint and Word, with Outlook in beta.
- Claude Science: a desktop app for scientific research. In beta.
One subscription covers all of them, at different usage levels depending on your plan.
"Claude Cowork is now just Claude" 🔀
That's Anthropic's own wording, and it's on the pricing page right now.
Until recently there were two separate modes: Chat for conversation and Cowork for handing off real work. Cowork is where you describe the result you want, step away, and come back to finished files. Anthropic describes it as bringing "Claude Code's agentic capabilities to knowledge work beyond coding."
Those two modes are now being merged into one conversation. Here are the facts as of today:
- It's "rolling out gradually, starting with Pro and Max plans on web, desktop, and mobile."
- Team and Enterprise "keep chat and Claude Cowork as they are today" for now.
- Once your account moves over, "you can't switch back to separate 'Chat' and 'Cowork' options."
- In the merged experience, there's no web search toggle. "Claude searches the web when it helps."
So if your screen doesn't match a tutorial you're watching, this is probably why. Different accounts are seeing different versions this month.
Getting started on every platform 📲
- Web: claude.ai. Nothing to install.
- Desktop (Mac, Windows, Linux beta): install it. It's the only place Claude can work with files on your computer, use its built-in browser, or operate your computer (computer use). The minimum versions are macOS 11, Windows 10, and Ubuntu 22.04 or Debian 12. On Free, the desktop app gives you chat only. Claude Code and Cowork start at Pro.
- Mobile: iOS and iPadOS 18 or later, Android 8 or later.
- Chrome: the Claude in Chrome extension, on paid plans. It doesn't work in other Chromium browsers or on mobile.
- Excel, PowerPoint, Word and Outlook: add-ins, on paid plans.
- Terminal and code editors: Claude Code works in the terminal and in VS Code, Cursor and JetBrains.
The plans, and what you get 💳
Prices checked on 23 September 2026. All prices are in US dollars.
| Plan | Price | Who it's for |
|---|---|---|
| Free | $0 | Trying it out. Sonnet and Haiku only, up to 5 projects |
| Pro | $20/mo, or $17/mo billed annually ($200 upfront) | Most people who use it for work |
| Max 5× | $100/mo | Heavy daily users |
| Max 20× | $200/mo | People who live in Claude Code or run agents all day |
| Team: Standard seat | $20/seat/mo annually · $25 monthly | Teams of 2 to 150 |
| Team: Premium seat | $100/seat/mo annually · $125 monthly | Heavy users on a team, with 5× Standard usage |
| Enterprise | $20/seat/mo annually plus usage at API rates | Minimum 20 seats self-serve, 50 through sales |
Pro has annual billing and Max doesn't. Max is "available as a monthly subscription only."
Enterprise charges you in two parts. You pay a seat fee, then you pay for what people actually use at API rates. Anthropic says "no plan or seat-level usage limits" apply on usage-based plans, so the bill depends on how much people use it. Budget for that.
Web prices can differ from app-store prices. Anthropic says "Mobile pricing may vary." If you subscribe through an app store, you may pay more than the web price.
What's included, from the pricing page:
- Free includes Sonnet and Haiku (no Opus, no Fable), up to 5 projects, web search, memory, skills, connectors and artifacts.
- Pro adds Opus, Claude Code, Cowork, Claude Design, Slides and Docs, Claude Science, Claude in Chrome, Microsoft 365, Research, and at least 5× the Free usage per session.
- Max gives you 5× or 20× Pro's usage, and Fable is included in your plan (see the models section).
How usage limits work ⏱
Claude doesn't give you a message count. It gives you a usage budget, and there are two limits running at the same time:
- 1A five-hour window. "Every plan has usage limits that reset on a rolling five-hour session window."
- 2A weekly limit on paid plans. It "resets at a fixed time each week that is assigned to your account."
- Your whole account shares one pool. Web, desktop, mobile and Claude Code all use the same budget. A long coding session in the afternoon means less chat that evening.
- Longer conversations and bigger models use the budget faster. A long thread on Opus uses much more than short questions on Sonnet.
- When you hit a limit, you have three options. You can "wait for it to reset, move to a higher plan, or, on paid plans, turn on usage credits to keep working at standard API rates." Usage credits are prepaid, you can set a monthly cap, and there's a daily limit of $2,000.
Limits just went up. With Opus 5.5, Anthropic is "increasing five-hour usage limits on Pro, Max, Team, and seat-based Enterprise plans." Subscribers also get a limit reset they can save and use whenever they want — it refills your five-hour or weekly limit once.
On Team plans, every member has their own limit. A Standard seat gets 1.25× Pro's per-session usage, and a Premium seat gets 6.25×.
The models, explained properly 🧠
| Model | Released | What it's for |
|---|---|---|
| Claude Opus 5.5 | 22 Sept 2026 | The new default. Anthropic says to "start with Claude Opus 5.5 for most workloads." The pricing page calls it the "Daily driver for agentic coding and enterprise work." |
| Claude Fable 5.1 | 1 Sept 2026 | "For demanding reasoning and long-horizon agentic work." Anthropic's "most capable widely released model." |
| Claude Sonnet 5 | 30 June 2026 | "The best combination of speed and intelligence." |
| Claude Haiku 4.5 | 15 Oct 2025 | "The fastest model with near-frontier intelligence." |
Opus 5.5 vs Fable 5.1. Fable is the more capable model on paper, and it costs 2.5× Opus 5.5 on the API. But Anthropic says Opus 5.5 "performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5." For most people, Opus 5.5 is the one to use, and Fable is for the hardest jobs.
How Fable is paid for depends on your plan:
- Max (and Team and Enterprise Premium seats): Fable is part of your plan. "You can use up to 50% of your weekly usage limits on Fable models at no extra cost."
- Pro (and Team Standard seats): "Fable models run on pay-as-you-go usage credits from the start." In other words, Fable costs extra on Pro from the first message.
- Free: Fable isn't available.
Some guides say Fable is pay-as-you-go on both Pro and Max. That's wrong for Max.
More new models are coming. "Claude Sonnet 5.5 and Claude Haiku 5.5 will follow in the coming weeks." The model list is about to change again.
Retirements to know about: Sonnet 4.5 retires "not sooner than September 29, 2026" and Haiku 4.5 "not sooner than October 15, 2026." Opus 5 became a legacy model on 22 September when Opus 5.5 took over. Opus 4.1, Opus 4 and Sonnet 4 have already retired.
Knowledge cutoffs: June 2026 for Opus 5.5 and Fable 5.1, and January 2026 for Sonnet 5.
Effort and thinking 🎚
Every current model thinks before it answers. What you control is how hard it thinks.
Effort levels: Low · Medium · High · Extra high · Max. Extra high and Max are available on Opus 5.5, Fable 5.1, Opus 5 and Sonnet 5. Haiku doesn't have an effort setting.
The default depends on the model:
- Opus 5.5 defaults to Medium.
- Fable 5.1 defaults to High in Claude Code, and to Medium in Claude on the web and in Cowork.
- Most other models default to High.
On the top models, you can't turn thinking off. Anthropic says "Thinking cannot be turned off in Claude when using Claude Opus 5.5, Claude Fable 5.1, or Claude Opus 5." The only thing you can adjust is the effort level.
How to choose: leave it on the default for everyday work. Raise it to High or Extra high when getting it wrong would be expensive — a contract, a financial model, an important decision. Keep Max for the hardest problems, because higher effort uses your usage budget faster.
Context windows: how much Claude can hold at once 📏
The context window is how much Claude can read and remember within a single conversation. In the Claude apps, on paid plans:
| Context window | Models |
|---|---|
| 1M tokens | Fable 5.1 · Opus 5.5 · Opus 5 · Sonnet 5 |
| 500K | Fable 5 · Opus 4.8 · Opus 4.7 · Opus 4.6 · Sonnet 4.6 |
| 200K | Every other model |
Anthropic's pages disagree here. The Team plan help article still says "200k context window." The context-window article gives the figures above. The Team article looks out of date, but if you're on Team, test it yourself before you rely on the bigger number.
A million tokens is roughly several full-length books. You can give it a year of meeting notes and ask what changed.
On the API, Fable 5.1, Opus 5.5 and Sonnet 5 have 1M tokens, and Haiku 4.5 has 200K. There's no extra charge for long inputs.
No ads 🚫
Short section, because the answer is simple. On 4 February 2026 Anthropic wrote: "Claude will remain ad-free. Our users won't see 'sponsored' links adjacent to their conversations with Claude; nor will Claude's responses be influenced by advertisers."
That applies to every plan, including Free. ChatGPT, by comparison, now shows ads on its free and cheapest plans.
To be fair, the same post also says "Should we need to revisit this approach, we'll be transparent." So it's a stated policy that could change, not a permanent guarantee.
Cowork: handing off real work 🛠
Here's what's available in Cowork, whether yours still shows it as a separate mode or it's already been merged in:
- Long-running tasks that keep going while you do something else
- Sub-agents: Claude splits a big job into parts and works on them in parallel
- Real files as the output: spreadsheets, documents and decks
- A built-in browser in the desktop app
- Computer use, where Claude operates your computer. It's "in beta on Pro and Max plans" on macOS and Windows
- Scheduled tasks, projects, plugins and connectors
Plans: Pro, Max, Team and Enterprise. Not Free.
Where it runs: the desktop app on Mac and Windows is generally available. Web and mobile are in beta. Cloud sessions run "on Anthropic's servers, in an isolated environment" and are also in beta.
Anything that touches your local files, your browser or your computer still needs the desktop app "open and connected."
Two prompts to try:
Go through the invoices in this folder, pull the supplier, date,amount and due date from each one, and build me a spreadsheetsorted by due date. Flag anything overdue in red.
Read these three competitor websites and build me a one-pagecomparison: pricing, who they target, what they claim, and onegap none of them cover. Give it to me as a Word document.
Projects 📁
"Projects are available to all users, including those with free Claude accounts. Free users can create a maximum of five projects."
A project holds three things:
- Instructions that apply to every chat inside it
- Knowledge files, meaning reference material Claude can use in any of those chats
- All the chats you have about that piece of work
Paid plans only: when a project's files get too big for Claude to read all at once, Claude switches to searching them for the relevant parts instead. Sharing a project with teammates needs Team or Enterprise.
File limits: 30 MB per file in a project.
Tip: put your standing instructions — your style, your constraints, who you're writing for — in the project, not in each chat. Then every conversation starts already briefed.
Memory and incognito chats 🧩
Memory is on by default on Free, Pro and Max. On Team and Enterprise it's off by default, and an owner has to turn it on.
| Control | What it does |
|---|---|
| Settings → Memory | Lists everything Claude remembers as separate topics — edit or delete each one |
| Pause memory | Stops it collecting anything new |
| Reset memory | "Permanently deletes all memories" |
| Sensitive topics toggle | Off by default |
Incognito chats are available on every plan. They're "not used for training", but they are kept for 30 days. They don't work inside projects. In the merged experience, incognito chats open in the older chat view, where Claude can't create files or run code. So incognito is useful for privacy, but it's not the same as private, and it has fewer features.
Tip: ask Claude "what do you remember about me?" and delete anything that's wrong. It takes two minutes, and every later answer gets better.
Searching your past chats is on paid plans only.
Artifacts 🧱
When Claude builds something you can use — a small web page, a calculator, a chart or an interactive tool — it shows up as an artifact. That's a working object you can use, not just text on a screen.
- Plans: all of them, Free included. Code execution has to be on.
- Sharing: on Pro and Max you can choose "Only you" or "Anyone with the link." On Team and Enterprise you can also share with your whole organisation.
In the new Share dialog, "a Claude account is required" to open a shared artifact. If you want people without a Claude account to see it, check the sharing setting before you send the link.
Research and web search 🔎
Web search is on every plan, Free included. In the merged experience it just happens when it's useful, and there's no toggle.
Research is Claude's deep mode for complex questions. It runs many searches, reads through the results, and gives you a report with citations. It's "available for users with paid Claude plans," and web search needs to be on.
The habit that makes it work: tell it what decision the research is for. "Research standing desks" gets you a list. "I'm choosing between two standing desks for a small studio, what matters and which wins" gets you an answer.
Skills, connectors and plugins 🔌
These three are related, and they now live in one directory under Customize.
- Skills are saved instructions for a specific task. You can use them and make your own on every plan, including Free, as long as code execution is on. The one exception is recording a skill by showing Claude what to do — that's Pro, Max and Team only, and only in Cowork on Mac.
- Connectors link Claude to your apps — Google Drive, Slack, Notion — so it can read your data and take actions there. Available on all plans. Free users can add one custom connector.
- Plugins bundle everything together. "Each plugin bundles skills, connectors, and sub-agents into a single package." They're on paid plans only. They first launched on 30 January 2026 as a research preview.
Tip: install a plugin for your job before you write your own skills. Someone has usually packaged most of what you need already.
Scheduled tasks ⏰
Scheduled tasks let Claude do things on a schedule without you: "hourly, daily, weekly, on weekdays, or manually."
- Plans: Pro, Max, Team and Enterprise. Not Free.
- Runs while you're away: "Scheduled tasks run remotely, so they run on their cadence even when your computer is asleep or the Claude Desktop app is closed."
- The catch: "If a scheduled task requires local files or apps, it will only run locally." So a task that needs files on your computer only runs while your computer is on.
Good first tasks to set up: a Monday summary of your inbox, a daily check of competitors' pricing pages, a Friday wrap-up of what got done.
Note: Anthropic doesn't publish a maximum number of scheduled tasks.
Claude Code 💻
Anthropic's coding agent. It reads your code, writes and edits files, runs commands and tests, and works through a job step by step.
- Plans: Pro, Max, Team, Enterprise, or a Console (API) account. Not Free. It also works through Amazon Bedrock, Google Cloud and Microsoft Foundry.
- Where it runs: the terminal, VS Code, Cursor and JetBrains, the desktop app, the web, Slack, and the iOS and Android apps.
- Requirements: macOS 13 or later, Windows 10 1809 or later, or Ubuntu 20.04 or later, with at least 4 GB of RAM.
You don't need to be a developer to use it. "Write me a script that renames these 400 photos by date taken" is exactly the kind of job it's for.
Claude Design, Slides and Docs 🎨
These used to be one product. Now they're three, all in beta, paid plans only:
- Claude Design makes "designs, interactive prototypes, one-pagers, and other visual work." It launched on 17 April 2026. It no longer does slides — Anthropic now says "to make presentations, use Claude Slides."
- Claude Slides makes presentations.
- Claude Docs makes documents.
About the "design system" feature: some guides say Design applies your brand automatically. It only does that after someone has set up your design system in it. It doesn't work out your brand on its own.
On Enterprise, Design is off by default, and an admin has to turn it on.
Claude in Chrome 🌐
A Chrome extension that can read the page you're on, click, type, move between pages and fill forms.
- Paid plans only (Pro, Max, Team, Enterprise). It isn't available on Free, even for basic chat, despite what other guides say.
- It works in Chrome only. Not other Chromium browsers, and not mobile.
- The side panel is in beta.
The same safety rule applies here as anywhere: don't give it passwords or payment details to type in. Sign in yourself, then let it take over.
Microsoft 365: two different things 📊
- The add-ins. Claude runs inside Excel, PowerPoint and Word (with Outlook in beta). Anthropic says they're "now generally available on all paid plans." Paid plans only.
- The Microsoft 365 connector. This lets Claude read from SharePoint, OneDrive, Outlook and Teams. It's "available on all Claude plans," Free included. It needs a work Microsoft account, so personal @outlook.com accounts won't work.
People mix these two up. The add-ins put Claude inside Office. The connector lets Claude read your Microsoft data from inside Claude.
Claude Science 🔬
"Anthropic's desktop app for scientific research, currently in beta", on Pro, Max, Team and Enterprise. Claude writes and runs analysis code in an isolated sandbox, pulls data from scientific databases, connects to computing clusters, and saves every version of its results.
Anthropic's pages disagree on which computers it runs on. The help article says macOS and Linux. The product page says macOS, Windows and Linux. If you're on Windows, check before you plan around it.
Voice 🎙
Voice mode is in beta, on every plan, in the mobile app, the desktop app and the web. Fable models can't be used in voice mode.
Images: Claude doesn't generate them 🖼
This is the thing beginners most often get wrong. Anthropic says: "Claude doesn't generate photos or illustrations the way image-generation tools do."
- Reads images. Screenshots, photos, charts, handwriting, receipts.
- Builds visuals from code. Diagrams, charts, infographics and layouts in HTML or SVG.
If you want a photo or an illustration, use a dedicated image tool. Claude is very good at writing the prompt for that tool, though.
Files and uploads 📎
| Limit | Value |
|---|---|
| Per file, in a chat | 500 MB |
| Files per chat | 20 |
| Per file, in a project | 30 MB |
| Images | Up to 8000 × 8000 pixels |
| PDFs | Up to 1,000 pages. Charts and images are only read in PDFs of 100 pages or fewer |
That last point catches people. In a 400-page report, Claude reads the text but skips the charts and images. If the charts matter, upload those pages as a separate file.
Team and Enterprise 🏢
Team
- Seats cost $20 a month on annual billing or $25 monthly. Premium seats cost $100 annually or $125 monthly and get 5× the usage of a Standard seat.
- A Team plan needs at least 2 members and allows up to 150 seats.
- Every member has their own usage limit — one heavy user doesn't use up everyone else's budget.
- You get shared projects, admin controls, and no training on your data by default.
Enterprise
- $20 per seat per month on annual billing, plus usage at API rates.
- The minimum is 20 seats if you sign up yourself, or 50 if you go through Anthropic's sales team.
- Usage-based plans have "no plan or seat-level usage limits." That's powerful, but it means the bill depends on how much people use it — put a budget owner in charge.
- Keeping all AI processing in the US costs 1.1×.
Anthropic is simplifying Enterprise seats. The older Standard, Premium and Chat-only seats "are no longer available for new contracts" and move to a single Enterprise seat at renewal.
Privacy and data 🔒
The most important setting. On Free, Pro and Max, Anthropic's pricing page lists model training as "Opt-out." In other words, your chats can be used to train Anthropic's models unless you turn that off. The setting is in Settings → Privacy → "Help improve our AI models."
| Situation | Retention |
|---|---|
| If you allow training | Anthropic "may retain your data in a de-identified format for up to 5 years" |
| If you opt out | The usual 30-day retention applies |
| Deleted chats | Removed from Anthropic's systems within 30 days |
| Chats that broke usage policy | Kept for up to 2 years |
| Incognito chats | Not used for training, but kept for 30 days |
Team and Enterprise are different by default: "We will not use your chats or coding sessions to train our models, unless you choose to participate in our Development Partner Program." For a business, that alone can justify paying for Team seats instead of personal Pro accounts.
Prompting Claude well ✍️
Claude responds best to plain, specific briefs. You don't need tricks. Give it four things:
CONTEXT: who this is for, and what's already decided.TASK: the one thing you want, stated plainly.CONSTRAINTS: length, format, tone, anything it must not change.DONE LOOKS LIKE: what a good answer contains.
Four habits that make the biggest difference:
- 1Show it an example. A sample of the tone you want works better than any description of that tone.
- 2Ask it to disagree with you. "What's wrong with this plan?" gets you more than "Is this plan good?"
- 3Ask for the finished file. "Give me the spreadsheet" beats "tell me how to build the spreadsheet." Cowork can deliver the actual file.
- 4Raise the effort when getting it wrong would be expensive, instead of rewriting your prompt five times.
30 things to try this month 🔑
Your files and your computer (Cowork, desktop app)
Sort my Downloads folder into subfolders by file type and date.Show me the plan before you move anything.
Read every PDF in this folder and give me a one-line summary ofeach in a spreadsheet, with the filename in column A.
Find the duplicate photos in this folder and move them into aseparate folder called "duplicates". Don't delete anything.
Go through these receipts and build me an expense sheet bycategory and month, with totals.
Rename every file in this folder to the formatYYYY-MM-DD_short-description, using the date inside the document.
Writing and content
Here are three posts I wrote that performed well. Study the tone,then write five new ones on [topic] in the same voice.
Turn this messy voice-note transcript into a clean blog post.Keep my phrasing where it's good. Cut the rambling.
Rewrite this email so it's half the length, and it still makesthe one ask clearly.
Give me ten hooks for a video about [topic]. Each one undertwelve words, and none of them starting with a question.
Read this draft as a sceptical reader would, and tell me thethree places you'd stop reading.
Business and work
Build a spreadsheet that models my monthly revenue at threeprice points, with working formulas I can change.
Turn these meeting notes into decisions, owners and deadlines,in a table.
Here's my pricing page. Tell me what a first-time visitor wouldmisunderstand, and rewrite the parts that cause it.
Draft a polite but firm reply to this client who's asking forextra work outside the scope we agreed.
Make a 10-slide deck from this document for a client who hasten minutes. One idea per slide.
Learning and research
Explain [topic] to me like I'm smart but new to it. Then quizme with five questions, one at a time.
Research [decision] and tell me which option wins for someonein my situation: [situation]. Cite your sources.
Summarise this 80-page report in one page, then list anythingin it that contradicts its own summary.
Here's a contract. List every obligation I'm taking on and everydeadline, in plain language.
Make me a two-week study plan for [skill], 30 minutes a day,with a small project at the end.
Everyday life
Plan a 5-day trip to [place] for two people who like food andwalking, not museums. Rough budget [amount].
Here's what's in my fridge. Give me three dinners, anda shopping list for the week.
Help me compare these three phone plans for how I actually usemy phone: [usage].
Turn this list of chores into a weekly rota for four peoplethat's fair.
Here's a letter from [organisation]. Tell me what it's askingme to do, and by when.
Code (no experience needed)
Build me a simple web page that tracks my habits with checkboxes,one row per day.
Write a script that downloads every image from this list oflinks and names them in order.
This spreadsheet formula gives me an error. Fix it and explainwhat was wrong in one sentence.
Make a small calculator that tells me how much to charge for aproject, based on hours, rate and a margin I choose.
Look at this folder of code and explain what it does in plainEnglish, as if I'm the new owner of it.
Your first week 📅
- Day 1: set it up. Install the desktop app. Write your profile and instructions: who you are, what you do, and how you want answers. Check Settings → Privacy and decide on "Help improve our AI models."
- Day 2: make a project. Pick something you're already working on. Add instructions and two reference files. Do the rest of the week inside it.
- Day 3: ask for a file, not an answer. Give it one real job that ends in a spreadsheet, document or deck.
- Day 4: raise the effort. Ask the same hard question at Medium and then at Extra high, and compare the answers.
- Day 5: check its memory. Ask what it remembers about you, and fix or delete what's wrong.
- Day 6: automate one thing. Set up a scheduled task, like a weekly summary or a daily check. (Pro and above.)
- Day 7: find where it goes wrong. Give it a question you already know the answer to and see where it slips. That calibrates everything you do with it afterwards.
The honest bit 🙏
This page will go out of date. I checked everything against Anthropic's own documentation on 23 September 2026, the day after Opus 5.5 came out. Sonnet 5.5 and Haiku 5.5 are due within weeks, Chat and Cowork are still merging, and older models are about to retire. If you're reading this months from now, the structure will still hold, but check every specific number again.
Where Anthropic hasn't published something, I've said so. For example, the maximum number of scheduled tasks, and whether the training toggle starts switched on for new accounts. Anthropic's pricing page only says "Opt-out." A guide that gives you a firm answer on those is guessing.
Sources 🔗
Every fact on this page comes from Anthropic's own pages. I didn't use any third-party blogs.