AI Assistant Summary: Anthropic’s official documentation is comprehensive but sprawling. For someone learning Claude in 2026, twelve specific pages cover 95% of what you actually need: the Models Overview, Pricing, Quickstart, Prompt Engineering Guide, Tool Use, Extended Thinking, Computer Use, Prompt Caching, Batch API, Vision, the Cookbook on GitHub, and the Model Deprecations page. This guide curates those twelve, explains exactly when to read each one, and points to the canonical URL for every page. All URLs verified May 2026 against platform.claude.com/docs.
Anthropic publishes hundreds of pages of documentation across the Claude API, Claude.ai, the Console, and the official Cookbook on GitHub. New users routinely get lost trying to find the right page for the question they actually have. This curated list collapses that whole library into the 12 pages that are most useful most often. Each one is described in plain English, with a clear “read this if you…” trigger and the canonical Anthropic URL.
The 12 pages at a glance
| # | Page | Read it when you want to… |
|---|---|---|
| 1 | Models Overview | Pick between Opus, Sonnet, and Haiku |
| 2 | Pricing | Budget your API spend or pick a Claude.ai plan |
| 3 | Quickstart | Make your first API call in under 10 minutes |
| 4 | Prompt Engineering | Get better answers from any model |
| 5 | Tool Use (Function Calling) | Give Claude tools to call (web, code, your APIs) |
| 6 | Extended Thinking | Make Sonnet or Haiku reason step-by-step |
| 7 | Computer Use | Have Claude operate a desktop or browser |
| 8 | Prompt Caching | Cut API costs 50-90% on repeated context |
| 9 | Batch Processing | Get 50% off API costs for non-realtime work |
| 10 | Vision | Send Claude images and have it read them |
| 11 | Anthropic Cookbook | Copy a working code example for nearly any task |
| 12 | Model Deprecations | Avoid surprises when older models retire |
Below is a section for each page with details. The whole list is also available as a single bookmarks folder if you import them into your browser at once.
1-on-1 Coaching
Claude AI Crash Course
1-hour private video session with James. We walk through which Anthropic docs actually apply to your project, set up your Console, write your first prompts, and leave you with a working Claude integration. No technical background required.
Group Format
AI Workshops for Teams
Team-format workshops for businesses rolling Claude out to staff. We cover the Anthropic docs that matter for your stack, set up Console access, and walk your team through prompts that work for your actual workflows.
1. Models Overview
URL: platform.claude.com/docs/about-claude/models/overview
The single most important page on the whole site. It lists every current Claude model (Opus 4.7, Sonnet 4.6, Haiku 4.5), their pricing, context windows, knowledge cutoffs, and primary use cases — in one comparison table. Every other page on Anthropic’s docs assumes you know which model you’re using; this is where you learn the answer.
Read it when: You’re picking between Opus, Sonnet, and Haiku — or trying to decide whether to upgrade from an older model. The table at the top tells you almost everything in under 60 seconds. Our Opus vs Sonnet vs Haiku comparison goes deeper if you need the practitioner-friendly version.
2. Pricing
URL: claude.com/pricing
This is the consumer-facing pricing page covering both Claude.ai plans (Free, Pro, Max) and the API. For full API pricing including prompt caching rates, batch discounts, and per-model breakdowns, the deep-dive lives at platform.claude.com/docs/about-claude/pricing.
Read it when: You’re estimating costs, comparing tiers, or trying to decide if Claude.ai’s $17-20/month Pro plan covers your usage or if you need API access.
3. Quickstart (Get Started)
URL: platform.claude.com/docs/get-started
The fastest path from “I have an Anthropic account” to “I just made a successful API call.” Python and TypeScript examples, with a working “Hello, Claude” you can copy and paste. If you’ve never called the Claude API before, this is the first page you read after signing up.
Read it when: Day 1 of building with Claude. After this you can branch into Tool Use, Prompt Engineering, or whatever your project needs.
4. Prompt Engineering
URL: platform.claude.com/docs/build-with-claude/prompt-engineering
Anthropic’s own guide to writing prompts that work. Covers the standard techniques (be specific, give examples, structure your output) plus Claude-specific patterns like using XML tags to separate system content from user content. The most actionable single source for getting better answers from Claude.
Read it when: Your prompts aren’t giving you the results you expect. Or before you start, if you’re a careful person.
5. Tool Use (Function Calling)
URL: platform.claude.com/docs/build-with-claude/tool-use
How to give Claude tools to call — your APIs, search engines, code interpreters, databases. This is the foundation for agents: a Claude model that can take actions in the world rather than just respond with text. Includes the JSON schema for defining tools and patterns for handling tool-call results.
Read it when: You want Claude to do something beyond returning text — query your database, hit an API, run a calculation, browse the web. For most production AI applications past basic chat, this is where the real work starts.
6. Extended Thinking
URL: platform.claude.com/docs/build-with-claude/extended-thinking
The feature that lets Sonnet 4.6 and Haiku 4.5 produce visible step-by-step reasoning before their answer. You turn it on per request and the model exposes its “thinking” tokens. Useful for hard reasoning problems, math, and any task where you want to inspect the chain of thought rather than just trust the conclusion. Opus 4.7 uses Adaptive Thinking instead and doesn’t expose traces.
Read it when: You’re getting wrong answers on hard reasoning and want the model to slow down and show its work. Or when you specifically want to log/audit the model’s reasoning.
7. Computer Use
URL: platform.claude.com/docs/build-with-claude/computer-use
The feature that lets Claude control a virtual desktop — click, type, take screenshots, navigate UIs. This is a step beyond Tool Use: instead of calling your APIs, Claude is operating software the way a human would. Anthropic’s reference implementation walks you through the security model, the screenshot tool, and how to build a sandboxed environment.
Read it when: You want Claude to do something inside an existing GUI application — fill out a web form, navigate a desktop app, automate a workflow that doesn’t have an API. Our Claude Computer Use guide covers practical setup.
8. Prompt Caching
URL: platform.claude.com/docs/build-with-claude/prompt-caching
The single biggest cost-cutting feature on the Anthropic API. If you send the same large context across many requests (a knowledge base, a long system prompt, a codebase), prompt caching lets Anthropic store the processed context and charge you 10% of the normal input rate on cache hits. Real-world deployments cut their costs 50-90% with this one feature.
Read it when: You’re paying meaningful money on the Claude API and want to cut it. The implementation is a one-line change in most SDKs and the savings start on the very next request.
9. Batch Processing
URL: platform.claude.com/docs/build-with-claude/batch-processing
For workloads that don’t need an instant answer — classification jobs, content moderation, overnight summarization — Anthropic’s Batch API runs them at a flat 50% discount. You submit a batch (up to 100,000 requests at a time), Anthropic processes within 24 hours, you get the results back. Easiest API cost cut for non-realtime work.
Read it when: You’re running large background jobs and don’t need each result within seconds. Pair with prompt caching for compound savings.
10. Vision (Image Input)
URL: platform.claude.com/docs/build-with-claude/vision
How to send Claude images alongside text prompts. Covers the input format (base64 or URL), supported formats (PNG, JPEG, GIF, WebP), token-cost calculation, and patterns for OCR, screenshot analysis, document images, charts, and diagrams. All current Claude models support vision; this page tells you how to use it properly.
Read it when: Your application processes images — receipts, charts, screenshots, photos, scanned PDFs. Or when text extraction from images isn’t giving you what you need.
11. The Anthropic Cookbook
URL: github.com/anthropics/anthropic-cookbook
Not technically docs — it’s a GitHub repository of working code examples. Anthropic engineers and the community contribute Jupyter notebooks showing how to do specific tasks: retrieval-augmented generation, agent loops, multi-turn evaluation, citation handling, fine-grained tool use, vision pipelines, and dozens more. The cookbook is where you find working code for whatever you’re building.
Read it when: You know what you want to build but need a starting point. Clone the repo and search for the closest example to your use case. Almost every common pattern has an example.
12. Model Deprecations
URL: platform.claude.com/docs/about-claude/model-deprecations
The page nobody bookmarks until they regret not bookmarking it. Anthropic publishes its model lifecycle policy here: which models are deprecated, when they’ll retire, what to migrate to. As of May 2026, Claude Sonnet 4 and Opus 4 are retiring June 15, 2026 — if your code targets them, this is the page that tells you.
Read it when: You’re starting a new project (so you don’t pick a model about to retire). And once a quarter as a maintenance check on existing projects.
Pages worth knowing about but not in the top 12
If you need them, they’re here:
- Files API — upload documents once, reference them in many requests. Pairs well with prompt caching.
- Embeddings — not Anthropic-native (they recommend Voyage AI), but the docs explain how to integrate.
- Streaming — how to stream responses token-by-token instead of waiting for the full answer.
- Service Tiers — covers the Priority Tier for guaranteed throughput on production workloads.
- Reduce Hallucinations — Anthropic’s own patterns for citing sources and admitting uncertainty.
- Evaluation framework — how to measure Claude’s quality on your specific task.
- Transparency Hub — Anthropic’s policies, model cards, safety research summaries.
- Model Context Protocol — the open standard Anthropic created for connecting tools to LLMs.
How to actually use this list
Three workflows that have worked well:
- Bookmark folder. Save all 12 URLs to a single folder in your browser. When you have a Claude question, open the folder and skim the titles — one of them probably answers it.
- Cmd-K / quick-launch app. If you use Raycast, Alfred, Arc Search, or similar, create a snippet that lists these 12 with keyboard shortcuts. Searching “anthropic prompt” should drop you on the right page in two keystrokes.
- Print and pin. The 12-page list fits on one sheet. People who work next to Claude every day sometimes find a physical reference faster than another browser tab.
The Anthropic documentation is still growing, but these 12 cover almost every situation a beginner-to-intermediate Claude user runs into. The exotic stuff — finetuning, multimodal embeddings, region-specific compliance — is rare enough that you can find it on demand.
Frequently asked questions
Why does Anthropic’s docs URL keep changing?
The official docs moved from docs.anthropic.com to platform.claude.com/docs in 2025 as Anthropic rebranded the developer surface area under the Claude name. Old URLs 301-redirect to the new ones, so existing bookmarks still work. The new URLs are stable and worth updating any links you maintain.
Is there an offline version of the Anthropic docs?
Not officially. The docs site is generated, and Anthropic doesn’t publish a downloadable bundle. If you need offline access, third-party tools like gitingest can scrape the docs into a single file you can store locally or pass to Claude itself as context.
Where do I report errors in Anthropic’s documentation?
Each docs page has a feedback widget at the bottom (“Was this helpful?”). For substantive issues, you can also file a GitHub issue in the relevant SDK repository or contact Anthropic support through the support portal. They’ve been responsive to errata in our experience.
Are the SDK reference pages worth bookmarking too?
Only if you’re actively coding. The Messages API reference is the canonical surface, and the Python SDK reference is the most-used. They’re more useful as a Cmd-F target when debugging than as something you read end-to-end.
What about the Anthropic blog?
The Anthropic newsroom is where announcements land (new models, policy changes, partnerships). Worth subscribing to via RSS or following on social. It complements but doesn’t replace the documentation — news posts often link to the docs page that contains the actual reference.
Does Anthropic have official video tutorials?
Yes — the Anthropic Academy hosts free Claude courses. There’s also the official YouTube channel. For getting started, the Academy is a faster on-ramp than the docs alone.
How often does the documentation change?
Constantly — Anthropic updates the docs alongside product releases (every 2-4 weeks), and pricing/model-overview pages are revised whenever a new model ships. For consequential changes, follow our daily newsletter — we summarize what changed and why it matters.
Stay current on Anthropic releases
Get the day each new Anthropic doc lands
Free daily newsletter — one story, one tool, one tip. Plain English, no jargon. One issue per day.
Free forever. Unsubscribe anytime.
You may also like
- Anthropic Academy: Free Courses to Learn Claude
- Claude Opus vs Sonnet vs Haiku: Which to Use in 2026
- How to Use Claude AI: Complete Beginner’s Guide
- Claude API for Beginners: Build Your First AI App
- Claude Computer Use: Complete Guide
- Claude Skills: Build Custom AI Workflows
- What’s New in Claude 2026
- Claude Max vs Pro vs Free: Plan Comparison
- What Are AI Plugins?
Sources
- Anthropic. “Documentation index.” platform.claude.com/docs (verified May 14, 2026)
- Anthropic. “Models overview.” platform.claude.com/docs/about-claude/models/overview
- Anthropic. “Pricing.” claude.com/pricing
- Anthropic. “Cookbook.” github.com/anthropics/anthropic-cookbook
- Anthropic. “Model deprecations.” platform.claude.com/docs/about-claude/model-deprecations
Two ways to go further
The AI Prompt Library
1,000+ ready-to-use prompts for Claude, ChatGPT, and Gemini. Stop staring at a blank box.
Get it for $39 →2-Hour Live AI Crash Course
A private, beginner-friendly session across Claude, ChatGPT, Gemini, and the wider landscape.
Book for $125 →