Claude API Keys: Getting a Key, Pricing & Usage (2026)

How to get a Claude API key, per-model pricing (Haiku/Sonnet/Opus), billing and credits, cost savings (caching/batch), and key security — all in one guide.

🌐 This article was machine-translated and may contain inaccuracies. Refer to the Korean original if in doubt.

To call Claude from code, you need an API key. The API is completely separate from claude.ai subscriptions (like Pro) — it's pay-as-you-go, billed per token. This guide covers where and how to get a key, what it costs, and how to keep your key safe. (For the actual first call, see the API getting-started guide.)

Claude API pricing by model (per MTok) Model Input Output Haiku 4.5 $1 $5 Sonnet 4.6 $3 $15 Opus 4.7-4.8 $5 $25 As of May 2026. Output is ~5x input. Prompt caching up to -90%, Batch API -50%. Prices may change.

Subscription vs API

This trips people up most. Claude Pro ($20/mo) does not include API usage. The API isn't a monthly subscription — it's pay-as-you-go, billed by the input tokens you send and output tokens you receive. Call nothing and you owe nothing.

How to get an API key

  1. Sign up / log in at platform.claude.com (the developer console). New accounts usually get a small amount of free credit, though the amount and terms can change.
  2. Add a payment method and top up credits. New accounts start on prepaid credits with optional auto-reload; higher-usage accounts can switch to monthly invoicing.
  3. Create a key under API Keys. The key value is often shown only once, so store it somewhere safe immediately.

Pricing by model

Pricing bills input tokens (your prompt/context) and output tokens (Claude's reply) separately, and output is ~5x input across all current models. The three recommended models as of May 2026 are in the table above.

  • Haiku 4.5 ($1/$5): cheapest; great for classification, extraction, summarization, high-volume work.
  • Sonnet 4.6 ($3/$15): best price-to-quality balance for general tasks.
  • Opus 4.7-4.8 ($5/$25): most capable; for complex coding, reasoning, and agents (Opus 4.8 is the latest).

Cutting costs

  • Prompt caching: cache long, repeated prefixes for up to 90% off cached input.
  • Batch API: process non-urgent work in bulk for a flat 50% discount.
  • Use the right model. Sending simple tasks to Opus is a common source of overspend. Combining caching and batch helps a lot.

Rate limits

The API has per-minute request and token limits that increase by tier based on your payment and usage history. Check the limits page in the console for exact numbers. Exceeding a limit returns a 429 error.

Key security — don't skip this

  • Keep keys in environment variables, e.g. export ANTHROPIC_API_KEY="sk-ant-..."
  • Never commit keys to code or Git. A public repo can get a key stolen instantly.
  • Rotate keys periodically; if you suspect exposure, revoke and reissue in the console immediately.
  • Never put keys in front-end (browser) code — call only from your server.

Next steps

Once you have a key, see the API getting-started guide for the first request/response. For billing or key issues, use the platform.claude.com console and official docs.

Disclaimer: Pricing, free credits, and limit policies change often. The figures here (Haiku $1/$5, Sonnet $3/$15, Opus $5/$25, caching -90%, batch -50%) reflect Anthropic's official pricing at the time of writing (May 2026); check platform.claude.com for the latest. This site is not affiliated with Anthropic.

Keep reading

Have a question or want to share how you use Claude?

Join the community to share tips with other users, or explore more guides.