How do I authenticate with the Claude API?

Set the ANTHROPIC_API_KEY environment variable and the SDK reads it automatically when you create a client. Pass api_key directly to the constructor only if you manage multiple keys or a secrets manager.

Which Claude model should I use by default?

Start with claude-opus-4-8 for most new integrations — it balances reasoning quality with cost. Switch to claude-haiku-4-5 for high-volume, latency-sensitive tasks like tagging or routing, and reach for claude-sonnet-5 as a faster, cheaper middle ground.

What does stop_reason tell me about a response?

stop_reason explains why Claude stopped generating: end_turn means a normal finish, max_tokens means the response got cut off, tool_use means Claude called a tool, and refusal means the safety classifier declined the request. Always check it before reading content .

Why did my request fail before it even reached Claude?

The Messages API requires model , max_tokens , and messages on every request. Missing any of these — most often max_tokens — fails validation before the request is sent, so double-check your payload against the required fields first.

Build / AI / Claude API

Claude API Basics

Q: How do I build a multi-turn conversation with the Claude API?

Append each turn to the messages array in order, alternating user and assistant roles. Claude has no memory between calls, so you must resend the full history on every request. messages = [ { "role" : "user" , "content" : "What is a closure?" }, { "role" : "assistant" , "content" : "A closure captures variables from its enclosing scope." }, { "role" : "user" , "content" : "Give me a JavaScript example." }, ]

A quick reference for making your first Claude API call, picking a model, and reading the response.

TL;DR

Install the Anthropic SDK and set your ANTHROPIC_API_KEY variable.
Call client.messages.create() with a model, prompt, and max_tokens.
Check stop_reason before reading content to handle responses safely.

Jump to section

Installation and Auth
Your First Request
Reading the Response
Choosing a Model

Installation and Auth

pip install anthropic

Installs the Python SDK from PyPI so you can start calling Claude.

pip install anthropic

npm install @anthropic-ai/sdk

Installs the TypeScript and Node SDK for JavaScript projects.

npm install @anthropic-ai/sdk

ANTHROPIC_API_KEY

Stores your API key as an environment variable instead of hard-coding it.

export ANTHROPIC_API_KEY="sk-ant-..."

Your First Request

Anthropic()

Creates a client that reads the API key from the environment automatically.

import anthropic
client = anthropic.Anthropic()

messages.create()

Sends a prompt to Claude with a model, max_tokens, and a messages array.

message = client.messages.create(
    model="claude-opus-4-8",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Explain closures in one paragraph."}]
)

system parameter

Gives Claude a persistent role or persona for the whole conversation.

message = client.messages.create(
    model="claude-opus-4-8",
    max_tokens=1024,
    system="You are a concise technical writer.",
    messages=[{"role": "user", "content": "Explain closures."}]
)

max_tokens

Caps the response length; short answers need less, long docs need more.

# Short answers: 256-512  |  Long docs: 4096-8192  |  Max: 128000

Reading the Response

content[0].text

Reads the text of Claude's reply from the response object.

print(message.content[0].text)

stop_reason

Explains why generation stopped: end_turn, max_tokens, tool_use, or refusal.

if message.stop_reason == "end_turn":
    print(message.content[0].text)

usage

Reports input and output token counts so you can track cost.

print(message.usage.input_tokens)
print(message.usage.output_tokens)

message.model

Confirms which model actually served the request.

print(message.model)  # e.g. "claude-opus-4-8"

Choosing a Model

Claude Opus 4.8

Best for hard reasoning and agentic tasks; the default for most new integrations.

claude-opus-4-8

Claude Sonnet 5

Near-Opus quality on coding and agentic work at a lower cost.

claude-sonnet-5

Claude Haiku 4.5

Fastest and cheapest option for high-volume tasks like tagging or routing.

claude-haiku-4-5

Claude Fable 5

Anthropic's most capable model for the hardest, longest-horizon tasks.

claude-fable-5

Tips

Pin an exact model ID like claude-opus-4-8 in production so a later model update never silently changes your app's behavior.
Set max_tokens generously for long or open-ended replies, since Claude stops the moment it reaches that limit, even mid-sentence.

Warnings

max_tokens is required on every request — omitting it raises a validation error before Claude even sees your prompt.
Never hard-code an API key in source control; load it from an environment variable or a secrets manager instead.

In Practice

Sends a single prompt to Claude, checks the stop reason, and prints the response text and token usage.

Creating the client once and reusing it avoids re-reading credentials on every call.
Passing max_tokens and model explicitly keeps the request valid and predictable in production.
Checking stop_reason before touching content prevents crashes when a response is truncated or refused.
Printing usage after every call makes token cost visible instead of a surprise on the bill.

import anthropic

client = anthropic.Anthropic()

message = client.messages.create(
    model="claude-opus-4-8",
    max_tokens=1024,
    system="You are a concise technical writer.",
    messages=[{"role": "user", "content": "Explain what a closure is, in two sentences."}]
)

if message.stop_reason == "end_turn":
    print(message.content[0].text)
elif message.stop_reason == "max_tokens":
    print("Response was cut off - increase max_tokens")
elif message.stop_reason == "refusal":
    print("Request declined by safety classifier")

print(f"Tokens used: {message.usage.input_tokens} in / {message.usage.output_tokens} out")

FAQ

Append each turn to the messages array in order, alternating user and assistant roles. Claude has no memory between calls, so you must resend the full history on every request.

messages = [
    {"role": "user", "content": "What is a closure?"},
    {"role": "assistant", "content": "A closure captures variables from its enclosing scope."},
    {"role": "user", "content": "Give me a JavaScript example."},
]