The Complete Claude Guide for 2026: Models, Plans, Claude Code and How to Use It Well

This Claude guide is the page we wish we had when we started building our apps with Claude. It covers everything in one place: what Claude is, which models and plans exist in October 2026, where you can use it, what Claude Code does, how to prompt it well, and what it all costs.

Claude changed a lot in 2026. Three new top models arrived in September alone, the plans were reshuffled, and Claude Code grew from a terminal tool into something you can run from your phone. If you last looked at Claude a year ago, most of what you know is out of date.

At OneClickTool we build Mac apps, iPhone apps and this website with Claude every day. This guide is written from that experience, with every price and date checked against Anthropic's own pages.

The short answer

• Claude is Anthropic's family of AI models and the apps around them: chat on web, desktop and mobile, Claude Code for building software, and the Claude API for developers.
• Four current models: Haiku 4.5 (small, cheap), Sonnet 5.5 (fast everyday), Opus 5.5 (the default for complex work) and Fable 5.1 (the most capable, and most expensive).
• Plans: Free (Sonnet and Haiku), Pro at $20 per month (adds Opus, Fable and Claude Code), Max from $100 per month (much more usage).
• Best at: long documents, careful writing, analysis and agentic coding. Not built for: generating images.
• Biggest tip: give Claude the whole job up front, including why it matters and how to check the result.

What is Claude?

Claude is an AI assistant made by Anthropic, an AI safety and research company. Under the hood it is a family of large language models. On top of those models sit several products:

• The Claude apps for chatting, writing, analysis and research, on the web, a desktop app and mobile apps.
• Claude Code, an agent that works directly in your software projects.
• The Claude API, which developers use to build Claude into their own products. The same models are also available through Amazon Web Services, Google Cloud and Microsoft.

When people say "Claude" they can mean any of these. The model is the brain; the products decide what it can touch and how you talk to it.

Claude models explained (October 2026)

Prices come from Claude's pricing page, checked October 1, 2026. A "token" is roughly three-quarters of a word; a million tokens is several long books.

Opus 5.5 (released September 22, 2026) is the model most people should default to. Anthropic says it performs at the level of Fable 5.1 on most work while costing 60% less per token, and it runs more than 30% faster than Opus 5.

Sonnet 5.5 (September 28, 2026) is the fast, lower-cost complement. It is excellent for well-defined tasks and anything you run many times.

Fable 5.1 (September 1, 2026) is Anthropic's most capable widely released model. It shares its underlying model with Mythos 5.1, which is only available to vetted organizations, and it is best kept as an escalation step. Our explainer on what Claude Fable is covers its unusual history, including a brief ban in June 2026.

For a side-by-side with benchmarks and a decision flow, read Claude Opus vs Fable vs Sonnet.

Claude in 2026: a quick timeline

A lot happened this year, and it explains why older articles about Claude feel wrong today:

The practical takeaway: the "best" Claude model changed three times in one month. Before you follow any advice about models, check its date.

Claude plans explained

A few things the table hides:

• Every plan has usage limits. They reset in rolling windows (commonly five hours) plus weekly caps. Bigger models use your allowance faster.
• Fable on Pro and Max draws from half of your weekly limit, according to the pricing page. Use it deliberately.
• Pro is the best value for most people who use AI for work: $20 gets you Opus, Fable and Claude Code. Move to Max only when you hit limits every week.

Where you can use Claude

On the web and desktop. The main chat app. You can upload files and images, organize work into Projects (a space with its own files and instructions), and ask Claude to build small charts, diagrams and interactive pages right in the conversation.

On your phone. The iPhone and Android apps include chat, and voice mode lets you talk to Claude hands-free. You can also check on Claude Code tasks from your phone.

In your code: Claude Code. More on this below.

Through the API. For developers building their own products. You pay per token, and you choose the model and effort level for each request.

What Claude does not do: generate photos or illustrations. It reads images well, but for image creation you need another tool. If images and voice are central to your day, read our ChatGPT vs Claude comparison.

Claude Code: building software with Claude

Claude Code is where Claude does the most real work for us. In Anthropic's words, it "reads your codebase, edits files, runs commands, and integrates with your development tools."

Where it runs: the terminal, VS Code and JetBrains, a desktop app, and the browser at claude.ai/code. The same project settings carry across all of them.

What makes it useful:

• Project memory. A CLAUDE.md file in your project tells Claude your stack, commands and rules at the start of every session. It can also read an AGENTS.md file written for other agents.
• Skills. Package a repeatable workflow, like reviewing a pull request or writing a blog post, as a command your whole team can use.
• Hooks. Run your own scripts before or after Claude acts, such as formatting every file it edits.
• Subagents. A lead agent splits a big task, runs helpers in parallel and merges the results.
• MCP connections. Link Claude to tools like Google Drive, Jira or Slack through the Model Context Protocol.
• Routines. Schedule Claude to run in the cloud, for example a weekly dependency check, even when your computer is off.

What it costs: Claude Code needs Pro, Max, Team or Enterprise, or API credits. The Free plan does not include it.

How it compares: OpenAI's Codex is the main alternative, cheaper to start and cheaper per token. We compare them honestly in Claude Code vs Codex.

From our own work: this blog, our Mac apps and our iPhone apps are built with Claude Code. The biggest lesson is that a short, precise CLAUDE.md does more for quality than any clever prompt. Tell it your commands, your folder layout and the mistakes you do not want repeated.

How to use Claude well: prompting that works

Claude's current models follow instructions closely. That is good news: clear instructions get clear results. It also means vague instructions get generic ones.

1. Say what "done" looks like. "Write a 150-word product description for a Mac app that cleans the Downloads folder, for people who are not technical" beats "write a description."

2. Give the context it cannot guess. Audience, constraints, examples of what you like, the files that matter. Attach documents instead of summarizing them.

3. Explain why. "Keep it under 170 characters because the App Store cuts it off" works better than "KEEP IT SHORT." Reasons help Claude handle cases you did not think of.

4. Ask for a format. Bullet list, table, a draft with three options, JSON. Say it.

5. Tell it how to check its work. For code, name the test command. For writing, give the checklist you would use. Claude is much better when it can verify.

6. For big tasks, ask for a plan first. Read it, fix the misunderstanding in one line, then let it work. It is far cheaper than fixing the wrong result.

7. Raise effort before switching models. On Opus 5.5 the default effort is medium. If an answer feels shallow, try high effort before paying for Fable. Our guide on how to use Claude Opus 5.5 effectively has the full effort ladder and ten more tips.

What Claude costs in practice

On a subscription, you pay a flat monthly fee and work within usage limits. For most individuals, $20 per month for Pro is the whole cost.

On the API, you pay per token. A quick way to estimate: a job that writes 100,000 output tokens costs about $0.50 on Haiku 4.5, $1 on Sonnet 5.5, $2 on Opus 5.5 and $5 on Fable 5.1, before input tokens.

Two cost habits that matter more than model choice:

• Prompt caching. If you reuse the same long instructions or documents, cached reads are far cheaper: $0.20 per million tokens on Opus 5.5 and Sonnet 5.5.
• Right-sizing. Send routine, repeated jobs to Sonnet or Haiku, and save Opus and Fable for work that needs judgment.

Using the Claude API: your first request

If you are a developer, the API gives you the same models with full control over model choice, length and cost. The basic steps:

1. Create an account on the Claude developer platform and add a small amount of credit.
2. Create an API key and store it as an environment variable, never in your code or a public repository.
3. Send a request to the Messages API with a model ID such as claude-opus-5-5, claude-sonnet-5-5 or claude-haiku-4-5.

A minimal request looks like this:

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{"model": "claude-sonnet-5-5", "max_tokens": 1024,
       "messages": [{"role": "user", "content": "Summarize this in 3 bullets: ..."}]}'

Official SDKs exist for Python, TypeScript and other languages, and they handle retries and streaming for you. Three habits keep API bills low from day one: start on Sonnet and move up only when needed, set a sensible max_tokens, and cache long instructions you send with every request.

Claude for different jobs

For picking the other tools around Claude, see our roundup of the best AI productivity tools.

Real examples: how we use Claude at a small studio

Concrete beats abstract, so here is how Claude fits into an ordinary week at OneClickTool:

• Building features. Claude Code with Opus 5.5 writes a new screen in one of our apps across the view, data and test files. We ask for a plan first, approve it, then review the diff and run the app before merging.
• Store listings. Claude drafts app descriptions, subtitles and keyword lists to fit each store's character limits, and a script counts the characters. It is a well-defined job, which is exactly where Sonnet 5.5 is enough.
• Research before a new app. Claude reads competitor reviews and pricing pages we collect, and summarizes what users complain about most. We check every number it quotes against the source.
• This blog. Claude drafts from our outline and sources, and a checker script rejects posts that are too thin, lack sources or use filler phrases. Images are generated separately, with a different tool.
• Repetitive chores. Skills in Claude Code package the steps we repeat, such as publishing a product page, so they run the same way every time.

The pattern: the harder the judgment, the bigger the model and the more we review; the more routine the job, the smaller the model and the more we automate.

Privacy and safety basics

• Check your data settings in Claude's privacy controls before sharing anything sensitive, and follow your employer's rules for company data.
• Expect refusals or rerouting on some topics. Fable 5.1 sends flagged cybersecurity, biology and chemistry requests to a less capable model, and every model declines clearly harmful requests.
• Verify important facts. Claude can still be wrong with confidence. For numbers, laws, medical or financial decisions, check the source.
• Review before you ship. With Claude Code, read the diff and run tests before merging, and keep work on a branch.

Common mistakes new Claude users make

• Using Fable for everything. It is slower, pricier and not better on most tasks than Opus 5.5.
• Starting a new chat for every follow-up. Keep related work in one conversation or Project so Claude keeps the context.
• Pasting summaries instead of sources. Give Claude the real document; it handles long inputs well.
• Accepting the first draft. Ask for three options, or tell it exactly what to change. The second draft is usually much better.
• Ignoring limits until they hit. If you run out mid-week, move routine work to Sonnet before upgrading.

Getting started: a 20-minute plan

1. Sign up for the Free plan and try Sonnet on a real task from your week.
2. Create one Project with your key documents and a short instruction ("You are helping me write product copy for…").
3. Try one long task: upload a long report and ask for the three decisions it implies, with page references.
4. Upgrade to Pro only when you want Opus, Fable or Claude Code, or keep hitting the free limits.
5. If you build software, install Claude Code, write a ten-line CLAUDE.md, and give it one small, real issue.

The bottom line

Claude in 2026 is four models, a handful of plans and three ways to use it: chat, Claude Code and the API. For most people the right setup is simple: Pro at $20 a month, Opus 5.5 as the default, Sonnet 5.5 for routine work, and Fable 5.1 held in reserve. Give it the whole job, explain why, and tell it how to check itself.

Where to go next:

• Choosing a model: Claude Opus vs Fable vs Sonnet
• Getting better results: how to use Claude Opus 5.5 effectively
• Understanding the top model: what is Claude Fable
• Choosing an assistant: ChatGPT vs Claude
• Choosing a coding agent: Claude Code vs Codex