Get API keys for every major model and make your first call safely

1Know what the API gives you, and what your subscription doesn't cover

Use the API to run AI inside your own scripts, automations or product, to pick a cheaper model per task, and to get business data terms: Anthropic, OpenAI and xAI don't train on API traffic by default, and neither does Google on its paid tier. Instead of message caps, you pay per token, within rate limits.

Your chat subscription is a separate product. As of October 2026:

2Create an account and a key at each provider

Use a work email and two-factor sign-in. Create one key per app and name it after the job ("invoice-bot-dev"), so you can revoke one without breaking the rest.

• Google's free tier costs you data. Under the Gemini API terms, free-tier prompts can improve Google's products and be read by human reviewers, so send nothing confidential. If a key shows Unrestricted, hover over it and choose Add restrictions > Restrict to Gemini API only.
• OpenRouter passes through provider prices with no markup and charges 5.5% when you buy credits by card: one key and one bill for many models.
• OpenAI's Restricted permission limits a key to the endpoints a script needs. Use it for anything unattended.

3Cap spending before your first call

Prepaid credit is your first ceiling: with auto-reload off, a key can only spend what's in the account. Then set a monthly cap in each console.

4Store keys like passwords

• Master copy in a password manager, noting where each key is used.
• On your machine, a .env file listed in .gitignore before your first commit.
• Never in code, chats (AI chats included), screenshots, shared docs, or a website or app front end.
• One key per person and app. Anthropic and OpenAI both say don't share keys. Anthropic suggests rotating every 90 days.

ANTHROPIC_API_KEY=sk-ant-...
OPENAI_API_KEY=...
GEMINI_API_KEY=...
XAI_API_KEY=...
OPENROUTER_API_KEY=...

echo ".env" >> .gitignore
git check-ignore -v .env     # prints the matching rule if .env is ignored

Python reads it with python-dotenv (load_dotenv()). Node.js 20.6+ reads it with node --env-file=.env app.js.

5Make your first call from the terminal

Load .env into your shell, then run a call. Small models and a short prompt keep each one under a cent.

set -a; source .env; set +a      # loads the keys into this terminal only
curl https://api.anthropic.com/v1/messages \
  -H "content-type: application/json" \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-version: 2023-06-01" \
  -d '{
    "model": "claude-haiku-5-5",
    "max_tokens": 1000,
    "messages": [{"role": "user", "content": "Say hello in five words."}]
  }'

curl https://api.openai.com/v1/responses \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -d '{"model": "gpt-6-luna", "input": "Say hello in five words."}'

curl -X POST "https://generativelanguage.googleapis.com/v1beta/interactions" \
  -H "x-goog-api-key: $GEMINI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "gemini-3.8-flash", "input": "Say hello in five words."}'

The usage numbers in the reply are what you paid for. A 401 means the key is wrong or not loaded. A 429 means a rate limit or spend cap (on Claude, your own spend limit returns a 400 that says so). If a model name is rejected, check the provider's models page: names change often.

6Call from Python and JavaScript, including OpenAI-compatible APIs

from dotenv import load_dotenv
import anthropic

load_dotenv()                    # reads ANTHROPIC_API_KEY from .env
client = anthropic.Anthropic()

message = client.messages.create(
    model="claude-haiku-5-5",
    max_tokens=1000,
    messages=[{"role": "user", "content": "Give me three names for a bakery."}],
)
for block in message.content:
    if block.type == "text":
        print(block.text)
print(message.usage)             # the tokens you paid for

Many providers accept OpenAI's SDK with a different base URL and key (model IDs as of October 2026):

import os
from dotenv import load_dotenv
from openai import OpenAI

load_dotenv()
client = OpenAI(
    base_url="https://openrouter.ai/api/v1",
    api_key=os.environ["OPENROUTER_API_KEY"],
)
reply = client.chat.completions.create(
    model="anthropic/claude-sonnet-5.5",
    messages=[{"role": "user", "content": "Give me three names for a bakery."}],
)
print(reply.choices[0].message.content)
print(reply.usage)

// save as hello.mjs, run with: node --env-file=.env hello.mjs
import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://openrouter.ai/api/v1",
  apiKey: process.env.OPENROUTER_API_KEY,
});

const reply = await client.chat.completions.create({
  model: "anthropic/claude-sonnet-5.5",
  messages: [{ role: "user", content: "Give me three names for a bakery." }],
});
console.log(reply.choices[0].message.content);
console.log(reply.usage);

For Claude in JavaScript, use @anthropic-ai/sdk and the TypeScript tab of Anthropic's quickstart.

7Work out what it will cost

• You pay per million tokens (MTok), with separate input and output rates. Output costs more: 5x on Claude Sonnet 5.5.
• Tokens are word pieces. On Claude's current tokenizer, 1M tokens is roughly 555,000 words. Providers count differently, so trust the usage field.
• Input is everything you send each call, including the whole conversation so far. Long chats cost more per turn.
• Tools cost extra. Claude's web search is $10 per 1,000 searches plus tokens.

The formula: (input tokens × input price + output tokens × output price) ÷ 1,000,000.

Worked example (Anthropic prices as of October 2026): a script drafts replies to 1,000 customer emails a month. Each call sends 3,000 tokens (instructions plus the email) and gets 500 back.

Same job, a 100x spread. Start with the cheapest model that does it well, and check prices when you build: Google has scheduled increases for several Gemini models on December 1, 2026 and January 1, 2027. Current prices: Claude, OpenAI, Gemini, xAI, OpenRouter.

I'm planning to use an AI API for: [the job, e.g. "summarize 200 sales calls a month"].
Each run sends about [X] words of instructions and input, and gets about [Y] words back.
It will run about [N] times a month.
Current prices per million tokens, copied from the provider's pricing page: [paste rows].
Estimate tokens per run (show your conversion), cost per run and cost per month for the
two cheapest models that could do this well and one premium model. Then suggest a monthly
spend cap, and list anything that would push the real cost higher (long chat history,
retries, web search, images, agent loops).

What's next

Before you connect a key to email, files or the web, read Use AI safely and cheaply. Then build an agent with the Agent SDK, run it on a small server, or connect it to your tools with MCP.