Skip to main content

Claude cheatsheet

A one-page reference for using Claude effectively. For the full walkthrough, see the complete guide.

📖 Full guide: Claude →

Model tiers​

TierBest for
Haikufast, cheap, lightweight tasks
Sonnetbalanced — default workhorse
Opusdeep reasoning, complex multi-step work
Mythosfrontier-level, limited/gated

Higher tier = smarter but slower & pricier — pick by task complexity, not habit.

Where to access it​

  • Claude.ai — web/desktop/mobile chat
  • Claude API — for developers building apps
  • Claude Code — agentic coding tool
  • Claude Cowork — agentic non-dev knowledge work
  • Claude in Chrome / Excel / PowerPoint — task-specific agents
  • Claude Tag — Slack integration (@Claude)

Core capabilities​

Conversation & reasoning, writing/editing, coding/debugging, document & image analysis, web search/research, file creation, and Artifacts — standalone interactive content generated alongside the chat.

Prompting 101​

  • Be clear and specific — vague prompts get vague answers
  • Give it a role/context to calibrate tone or expertise
  • Show 1-2 examples (few-shot) for a specific format
  • Ask for step-by-step reasoning on hard problems
  • Iterate on the same prompt — don't restart cold
  • Specify format explicitly (length, structure, tone)

Good at / weaker at​

Good: ambiguous/open-ended reasoning, long-form writing, explaining complex topics, coding, summarizing large text.

Weaker: perfect recall of very recent events, guaranteed accuracy on niche claims, exact large-scale math (use code execution instead).

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 problems
  • Iterate instead of re-prompting from scratch
  • Let it use tools for code/data instead of mental math
  • Double-check high-stakes factual claims