Claude Capabilities Quick Reference — Claude Cert Academy

A practical guide to what Claude can do, which model to choose, and how to structure requests for best results.

Model Selection Guide

Choose based on the complexity of the task and your cost constraints.

• claude-haiku-4-5 — fastest and cheapest; use for classification, extraction, simple Q&A, routing tasks
• claude-sonnet-4-6 — best balance of capability and cost; use for coding, analysis, content generation, complex reasoning
• claude-opus-4-8 — highest capability; use for research synthesis, complex multi-step reasoning, nuanced judgment calls

Default to Sonnet for new projects. Switch to Haiku for high-volume simple tasks once you know the task profile. Use Opus only when Sonnet falls short.

What Claude Does Well

• Code generation and review across 20+ languages
• Document analysis, summarization, and information extraction
• Long-form writing: reports, emails, explanations, documentation
• Data transformation: JSON to CSV, reformatting, normalization
• Question answering over provided context (not trained knowledge)
• Translation across 50+ languages
• Structured output: JSON, YAML, tables, markdown

What Claude Does Not Do

• Browse the internet in real time (unless given a web_fetch tool)
• Access files on your filesystem (unless given a file-reading tool)
• Remember previous conversations (context is per-session only)
• Execute code unless a code execution environment is provided
• Make decisions with external state (payments, sending emails) without tools

Messages API Structure

Every Claude API call follows the same shape.

{
  "model": "claude-sonnet-4-6",
  "max_tokens": 1024,
  "system": "You are a helpful assistant.",
  "messages": [
    {"role": "user", "content": "What is the capital of France?"}
  ]
}

Key Parameters

• model — which Claude model to use (required)
• max_tokens — maximum tokens in the response (required)
• system — persistent instructions Claude follows throughout the conversation
• temperature — 0 to 1, controls randomness (default ~1); lower for factual tasks, higher for creative
• top_k / top_p — advanced sampling controls; rarely needed
• stop_sequences — list of strings where Claude stops generating

Prompt Patterns That Work

Role + Task + Format

State who Claude should act as, what to do, and how to return the result.

You are a senior software engineer reviewing code for security issues.
Review the following Python function and list any security vulnerabilities as a numbered list.
For each, explain the risk and the fix.

[code here]

Few-Shot Examples

Show Claude the pattern before asking it to apply it.

Convert these sentences to past tense:
Input: I walk to the store.
Output: I walked to the store.

Input: She runs every morning.
Output: She ran every morning.

Input: They build amazing products.
Output:

Chain of Thought

Ask Claude to reason step by step before answering, especially for math or logic problems.

Think through this step by step before giving your answer:
If a train travels at 60 mph for 2.5 hours, then at 80 mph for 1.5 hours, what is the total distance?

Context Window Limits

• claude-haiku-4-5: 200K tokens input, 64K output
• claude-sonnet-4-6: 1M tokens input, 64K output
• claude-opus-4-8: 1M tokens input, 128K output
• 1 token ≈ 0.75 words in English; a 10-page document ≈ 3,000 tokens

Common Pitfalls

• Vague instructions — be specific about format, length, and audience
• Assuming Claude knows your system — provide all relevant context
• Over-relying on training knowledge — for current facts, provide a document
• Not checking for hallucinations — Claude can be confidently wrong; verify factual claims
• Ignoring the system prompt — use it to set persistent rules, not just the first user message

Continue to Claude Cert Academy