Skip to main content

Claude Cheatsheet — Learn It in One Read

A quick-reference guide to understanding and using Claude (by Anthropic) effectively.

📋 Quick reference: Claude →
🧭 Mental modelThe prompt–model–output loopA prompt with context, goal, and constraints goes to a Claude model tier chosen for the task's complexity, which produces an output. That output is refined by iterating on the same prompt rather than starting over.Your promptcontext + goal +constraints + formatClaudeHaiku → Sonnet → Opuspicked by task complexityOutput / drafttreat as a first pass,not the final worditerate on the same prompt — don't restart from scratch

Everything else in this cheatsheet is detail on one loop: a well-specified prompt goes to the model tier that matches the task, comes back as a draft, and gets refined by iterating — not by re-prompting cold each time.

1. What Claude Is

Claude is a family of large language models (LLMs) built by Anthropic. It's a text-in, text-out (and now multimodal) AI assistant that can reason, write, code, analyze, and hold conversations. Think of it as a very capable, general-purpose "thinking partner" you interact with through natural language.


2. The Model Lineup (know the tiers)

Anthropic ships models in size/speed tiers, refreshed periodically:

TierBest for
HaikuFast, cheap, lightweight tasks (chat, simple Q&A, high-volume use)
SonnetBalanced — strong reasoning + good speed. Default workhorse for most tasks
OpusMost capable, slower/costlier — deep reasoning, complex multi-step work
Mythos (newest tier, above Opus)Frontier-level capability, limited/gated availability

Higher tier = smarter but slower & pricier. Pick based on task complexity, not habit.


3. Where You Can Access Claude

  • Claude.ai — web/desktop/mobile chat interface (what most people use)
  • Claude API — for developers building apps on top of Claude
  • Claude Code — agentic coding tool (terminal, desktop, or IDE) for delegating dev tasks
  • Claude Cowork — agentic tool for non-dev knowledge work
  • Claude in Chrome / Excel / PowerPoint — task-specific agents inside those apps
  • Claude Tag — Slack integration to tag @Claude into a channel

4. Core Capabilities

  • Conversation & reasoning — multi-turn dialogue, step-by-step problem solving
  • Writing — drafting, editing, summarizing, rewriting in different tones/styles
  • Coding — writing, debugging, explaining code across languages
  • Analysis — reading documents, spreadsheets, PDFs, images and reasoning over them
  • Search & research — can browse the web for current info when needed
  • File creation — docs, spreadsheets, slides, PDFs, code files
  • Artifacts — generates standalone interactive content (apps, diagrams, visualizations) alongside the chat

5. Prompting 101 (the highest-leverage skill)

Be clear and specific. Vague prompts get vague answers.

❌ "Write about marketing" ✅ "Write a 300-word LinkedIn post about why small SaaS companies should invest in SEO, casual tone, include one stat"

Give it a role/context when it helps calibrate tone or expertise:

"You're reviewing this code as a senior backend engineer focused on security."

Show examples (few-shot). If you want a specific format, show 1–2 examples of input → output.

Ask for step-by-step reasoning on hard problems:

"Think through this step by step before giving the final answer."

Iterate. Treat the first response as a draft — refine with follow-ups rather than starting over.

Specify format explicitly — length, structure, tone, whether you want bullet points, tables, or prose.

Break big tasks into smaller ones. Multi-step chains beat one giant ask.


6. Things Claude Is Good At vs. Not

Good at:

  • Reasoning through ambiguous, open-ended problems
  • Long-form writing and editing
  • Explaining complex topics simply
  • Coding, debugging, refactoring
  • Synthesizing/summarizing large amounts of text

Weaker at / be careful with:

  • Perfect recall of very recent events (has a knowledge cutoff — but can search the web when needed)
  • Guaranteed factual accuracy on niche/obscure claims — verify anything high-stakes
  • Math requiring exact precision on very large calculations (better to have it use code execution)
  • Won't help with malware, weapons uplift, or generating harmful/illegal content

7. Quick Best-Practice Checklist

  • State your goal, not just the topic
  • Give constraints (length, tone, audience, format)
  • Provide relevant context/documents upfront
  • Ask for reasoning on complex/ambiguous problems
  • Iterate instead of re-prompting from scratch
  • For code/data tasks, let it use tools (code execution, file creation) rather than doing math in its head
  • Double-check high-stakes factual or numeric claims

8. One-Line Summary

Claude is a reasoning-first AI assistant — the better and more specific your instructions, the better its output; treat it like a smart collaborator, not a search engine.