LLM APIs — First Call to Claude / GPT

• Install the Anthropic SDK; store your API key safely.
• Send a message with client.messages.create.
• Read the response object (.content[0].text, usage).
• Pick a model + max_tokens sensibly for cost.

pip install anthropic

# get a key from console.anthropic.com, then (don't hard-code it!):
export ANTHROPIC_API_KEY="sk-ant-..."     # macOS/Linux
setx ANTHROPIC_API_KEY "sk-ant-..."       # Windows

The first call

import anthropic

client = anthropic.Anthropic()   # reads ANTHROPIC_API_KEY from the environment

msg = client.messages.create(
    model="claude-haiku-4-5",        # fast + cheap; great for learning
    max_tokens=300,
    messages=[
        {"role": "user", "content": "Explain photosynthesis to a 10-year-old."}
    ],
)
print(msg.content[0].text)

The response object

print(msg.content[0].text)     # the reply text
print(msg.stop_reason)         # why it stopped (end_turn, max_tokens, ...)
print(msg.usage)               # input_tokens / output_tokens — your bill

You pay per token (input + output). max_tokens caps the reply length (and your cost). Check usage to learn how much each call costs.

Choosing a model (cost vs capability)

claude-haiku-4-5     fastest, cheapest — quizzes, simple tasks, learning
claude-sonnet-4-6    balanced — most apps
claude-opus-4-7      most capable — hard reasoning, complex tasks
(OpenAI's GPT models work the same way via the openai SDK.)

Keep your key out of code

# ✓ from environment (the SDK does this automatically)
client = anthropic.Anthropic()

# ✓ or via python-dotenv for local dev
from dotenv import load_dotenv
load_dotenv()        # reads .env (which is git-ignored!)

# ✗ NEVER:
client = anthropic.Anthropic(api_key="sk-ant-hardcoded")  # leaks in git!

Handle errors

try:
    msg = client.messages.create(model="claude-haiku-4-5",
                                 max_tokens=300,
                                 messages=[{"role": "user", "content": q}])
except anthropic.APIError as e:
    print(f"API error: {e}")

# explainer.py — ask Claude to explain anything simply
import anthropic

client = anthropic.Anthropic()

def explain(topic, age=12):
    msg = client.messages.create(
        model="claude-haiku-4-5",
        max_tokens=400,
        messages=[{
            "role": "user",
            "content": f"Explain '{topic}' to a {age}-year-old in 3 short "
                       f"sentences, with one everyday analogy."
        }],
    )
    return msg.content[0].text, msg.usage

text, usage = explain("how vaccines work")
print(text)
print(f"\n[tokens: in={usage.input_tokens}, out={usage.output_tokens}]")

Sample output

A vaccine is like showing your body a "wanted poster" of a germ. Your immune
system studies the poster and practises fighting that germ without you getting
sick. Then if the real germ ever shows up, your body already knows how to beat it.

[tokens: in=28, out=71]

Read the diff

The whole call is three lines; everything interesting is in the prompt (the next lessons are all about writing good prompts). Notice we logged token usage — get in the habit early, because it's your bill. For a learning app, Haiku keeps calls to fractions of a cent.

An LLM call is a REST request with auth, wrapped in an SDK: client.messages.create(model, max_tokens, messages) → .content[0].text. Keep the key in the environment, never in code. Pick the cheapest model that does the job (Haiku for learning). Watch usage — that's your bill. Next: shaping behaviour with system prompts.