API quickstart: your first call and what it costs

The API is the same Claude with a different meter and no interface. Use it when you want to automate a repeated task, build Claude into something, or when your usage is bursty enough that a subscription would bill idle weeks. Skip it if you just want to chat — a good client costs money and time, and Pro is simpler.

curl https://api.anthropic.com/v1/messages \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -H "content-type: application/json" \
  -d '{
    "model": "claude-sonnet-5",
    "max_tokens": 1024,
    "messages": [
      {"role": "user", "content": "In three bullets, when is an API cheaper than a chat subscription?"}
    ]
  }'

import anthropic

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

response = client.messages.create(
    model="claude-sonnet-5",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "In three bullets, when is an API cheaper than a chat subscription?"}
    ],
)

for block in response.content:
    if block.type == "text":
        print(block.text)

u = response.usage
print(f"input={u.input_tokens} output={u.output_tokens}")
# Sonnet 5 at the time of writing: $2 / 1M input, $10 / 1M output
print(f"cost ≈ ${u.input_tokens * 2 / 1e6 + u.output_tokens * 10 / 1e6:.5f}")

That call costs a fraction of a cent. The usage object is the whole cost-control story: every optimisation in lesson 8 is about making those two numbers smaller, or making the input tokens cheaper.

Two features to know before you build anything

• Prompt caching. Add cache_control to the stable part of a request — the system prompt, a big document, the tool list — and repeated calls read it from cache at about 10% of the normal input price. On a workload that re-sends the same 50,000-token document, that is the difference between a bill and a rounding error. Details: Anthropic's prompt caching docs.
• Message Batches. Send up to thousands of requests as one batch, get results within a day, pay half price. Anything that does not need an instant answer — nightly reports, bulk classification, evaluations — belongs here. Details: Batch processing docs.