Claude API Key: How to Get One, Is It Free, and Safe Buying

As of October 10, 2026, creating a Claude API key costs nothing, but the key only works while the Claude Console organization behind it has credit. Anthropic's pricing FAQ says new users "receive a small amount of free credits to test the API." Anthropic doesn't publish the amount. Third-party guides report about $5, and some say it depends on your country and a phone check. After that, you pay for usage with prepaid credits.

A Claude Pro subscription doesn't include any API usage. Max, Ultra (formerly Max 20x), and Team are the exception since October 7, 2026: they come with a monthly API credit that you claim into a Console organization you link. None of these plans gives you a key by itself. You still create the key in Console.

Don't buy a pre-generated or shared Claude API key from a marketplace or a "free Claude API" site. An API key is a credential, not a transferable balance. Whoever owns the organization behind it can see your usage, revoke the key, or spend the balance, and you can't do any of that yourself.

Is there a free Claude API key tier?

There is no ongoing free tier. There is a one-time new-user test credit, plus a few programs and plan perks that hand out API credit under conditions. The key itself is always free to create. What changes is who pays for the calls it makes.

The Max, Ultra, and Team amounts and conditions come from Anthropic's help article on monthly API credits for Max and Team plans. The research amounts come from the Claude Help Center pages on the External Researcher Access Program and Anthropic's AI for Science program. The $50,000-per-project figure comes from Anthropic's August 27, 2026 announcement on expanding support for scientists; the help article hadn't been updated to match as of October 10, 2026. For the startup program's status and the alternatives, see Claude for Startups Program: How to Apply, Team and $1K Paused.

The pricing FAQ also points enterprises to Sales for "extended trials for enterprise evaluation." Creating more keys doesn't multiply any of these balances. Every key in an organization spends from the same pool.

The last row matters because so many "free Claude API" offers exist. Some route you to free non-Claude models, such as OpenRouter's free models or NVIDIA NIM, under a Claude-sounding name. Others promise a million free Opus 5.5 tokens or extra credit for connecting a Telegram account. None of them gives you an Anthropic key. The model you actually get, what happens to your prompts, and how long the access lasts are all set by that site.

First decide whether you need raw API access

You need a Claude API key when your own code, SDK, CI job, agent, or bring-your-own-key app sends requests to Claude. You don't need one to use Claude on the web or in the apps.

A paid chat plan and the API are separate products. Anthropic's help article on why paid plans pay separately for the Claude API and Console puts it this way: "A paid Claude subscription enhances your chat experience, and the Claude API is billed separately." What Pro, Max, and Ultra multiply is chat and Claude Code usage, not API access. Claude Usage Limits by Plan: What 5x and 20x Actually Multiply covers that side.

Claude Code sits between the two. If it shows "Claude Max or Pro is required to connect to Claude Code. Sign up for a Max or Pro subscription to connect your account, or use your API key," it's asking you to pick one way to pay: sign in with a Pro, Max, or Ultra plan, or give it an API key that spends Console credits. The Max, Ultra, and Team monthly API credit doesn't pay for interactive Claude Code. Claude API vs Claude Code: Which One to Use and How You Pay helps you decide which one you need. If you're already in a terminal and unsure what's being billed, the Claude Code API billing versus subscription guide walks through it.

The official Console route is the reference path

For direct Anthropic API use, keep the account and the key under your control from the start. Anthropic's API access help names Claude Console as the place to create keys, manage users, and set up billing.

1. Check Anthropic's supported countries and regions for the person or company that will own the account. The United States, Japan, South Korea, Spain, and Taiwan are on the list. Mainland China, Hong Kong, and Russia aren't, so Anthropic doesn't offer direct API accounts there.
2. Sign in or create an account at platform.claude.com. Make sure your business or project owns the organization. Don't leave a contractor or seller as its only admin.
3. Use the new-user credit for a first test if your account shows one. To keep going, buy credits in Settings > Billing. You need the Admin or Billing role there.
4. Go to Settings → API keys and click Create key. Give it a name, pick an expiration, and set Linked account to yourself or to a service account. You can also scope it to one workspace.
5. Copy the key right away. It starts with sk-ant-, and Console shows it only once. If you lose it, create a new one. Anthropic's guide to getting a Claude API key shows the same steps.

If you bookmarked the old Console address, it still works. As of October 21, 2026, console.anthropic.com/settings/keys returns a 301 redirect to platform.claude.com/settings/keys, so "console.anthropic.com → API keys" and "Settings → API keys" on platform.claude.com are the same page.

The SDKs read the key from the ANTHROPIC_API_KEY environment variable. To use it in Claude Code, see Claude Code API Configuration: Keys, Settings, Models, and Gateways.

What the prepaid credit contract includes—and does not

Anthropic's API billing help says Claude API and Console playground usage are paid with prepaid credits. When the balance reaches zero, API calls stop working until you add credits or auto-reload tops up the balance.

These rules should shape how much you buy first:

• Credits expire one year after purchase, and the date can't be extended.
• All credit purchases are non-refundable.
• Anthropic charges only for successful calls. A request still counts if your client disconnects while it's running.
• Auto-reload buys more credit when the balance falls below a threshold you set. It prevents downtime, but it can also turn a runaway loop into repeated charges. Pair it with a spend limit.
• Max, Ultra, and Team monthly API credit is spent before any credit you bought. All keys in the linked organization share it.
• Monthly invoicing is available only through Anthropic's Sales team.

The help page doesn't state a minimum purchase, so check the amount Console offers you. If you're testing, buy small. Unused credit expires, and you can't get a refund.

Claude API tiers now: Evaluation, Start, Build, Scale, and spend caps

The old "Tier 1–4" chart with $5, $40, $200, and $400 deposits no longer applies. Anthropic's current rate limits documentation uses four tiers. Organizations are placed on a tier "automatically based on usage history and account standing," and new organizations "may start in the Evaluation tier, with limits below the standard limits."

Once you hit your tier's cap, requests return a 429 error with enforced_spend_limit_reached and no retry-after time until 00:00 UTC on the 1st of the next month. If you set your own spend limit below the cap, requests return a 400 error when you reach it.

On the Start tier, Opus 5.5, Sonnet 5.5, and Haiku 5.5 each get 1,000 requests per minute, 2,000,000 input tokens per minute, and 400,000 output tokens per minute. Fable 5.x models get 500,000 input and 100,000 output tokens per minute. Cache reads don't count toward the input limit. You can ask for more with Request tier increase on the Console Rate limits page. If you're already seeing 429s, Claude API Error: Rate Limit Reached: Find the Right Limit Before You Retry helps you figure out which limit you hit.

Update, November 10, 2026: Anthropic released Claude Fable 5.2 (claude-fable-5-2) on November 9 at $8 input and $40 output per million tokens; Fable 5.1 stays at $10 / $50. Cache reads on Fable 5.2 cost $0.20 per million tokens. A request with 2,000 input and 500 output tokens costs $0.036 on Fable 5.2, against $0.045 on Fable 5.1. To choose between the new model and Opus 5.5, see Claude Fable 5.2 vs Opus 5.5: When Fable High Beats Opus Max.

How much to buy first: Claude API prices per model

These are Anthropic's list prices per million tokens on the Claude API pricing page, as of October 10, 2026:

The Batch API halves these prices. US-only inference costs 1.1 times the list price. Opus 5.5 fast mode is $8 input and $40 output.

To size a first purchase, price one typical request. Cost = input tokens × input price ÷ 1,000,000 + output tokens × output price ÷ 1,000,000. For a request with 2,000 input tokens and 500 output tokens, at list prices without caching:

Output costs five times as much as input on every model, so long answers eat credit faster than long prompts. Replace the 2,000 and 500 with token counts from your own requests before you decide how much to buy. For Fable 5.1 in detail, see Claude Fable 5.1 API Pricing: What Changed and What You'll Pay. To compare against OpenAI, see the Claude versus OpenAI API cost guide.

Keep the key scoped, expiring, and replaceable

When you create a key, you pick one of three types:

• Personal key: acts as you, with your roles. It stops working when you leave the organization. Use it for your own development.
• Service account key: acts as a service account. Use it for CI, production services, and agents that several people share.
• Workspace key: a legacy type with no owner. It keeps working after its creator leaves, which is why Anthropic recommends the other two for new work.

Anthropic's authentication documentation lists the expiration choices: 3 hours, 1 day, 7 days, 30 days, a custom duration, or Never. You can't change the expiration after creation. An expired key returns a 401 authentication_error. Use a short-lived key for a quick test. If a production key has to live longer, keep it in a secrets manager and rotate it. Never put it in source code, a browser bundle, a screenshot, or a support message.

A key scoped to one workspace only works there. A key that isn't scoped needs the anthropic-workspace-id header on every request. The workspaces documentation explains how to split projects or environments and give each one lower spend and rate limits. Separate keys for development and production improve isolation, but they don't create separate free balances.

If you think a key has leaked, revoke it first and investigate second. On the API keys page, Disable can be undone and Delete is permanent. Anthropic's compromised-key guidance tells you to delete the affected key and create a new one. With a key you bought from someone else, you can't do either.

Apply the three-owner test before you pay

Before you pay anyone for Claude access, get three plain answers:

1. Who creates the credential and can revoke it?
2. Who owns the balance and bills each request?
3. Which exact endpoint accepts the key?

If the answers point to different or unnamed parties, you don't have a safe API arrangement yet. Use this table for any offer, whether it's Anthropic, a cloud provider, or a third-party site:

The first three rows tell you who you're dealing with. The last two tell you whether you can run it safely.

Why a cheap "$100 key" fails this test

Marketplace listings offer things like a "Claude API key with $100" or an "official direct" account preloaded with $1,000 of credit. Such a listing shows that someone offered a product. It doesn't show that you get the organization, sole control of the key, a balance that lasts, or anyone to ask for support or a refund. The key may work once and still leave every important control with the seller.

You don't need to prove that a seller is doing anything wrong to say no. This rule is enough: no account you own, no way to revoke the key yourself, and no usage records means no purchase.

Run one bounded call before production

Your first test should show that the key, the endpoint, and the billing record line up. It isn't a quality benchmark.

1. Set a spend limit for the organization or test workspace. Turn off auto-reload for the test, or set the reload amount small enough not to matter.
2. Create a test-only key with a short expiration, such as 3 hours. Don't reuse a production key.
3. Take the model ID, base URL, and headers from Anthropic's documentation. The direct endpoint is https://api.anthropic.com/v1/messages.
4. Send one prompt without sensitive data, such as "Reply with OK only," with a small max_tokens.
5. Find the request in Console usage and check the model, input and output tokens, and the charge.
6. Delete the test key and send the request again. It should fail authentication. If it still succeeds, stop and contact the provider before sending any real data.

A minimal request in Anthropic's documented format looks like this:

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-opus-5-5",
    "max_tokens": 16,
    "messages": [{"role": "user", "content": "Reply with OK only."}]
  }'

One passing call proves that this setup handled one request and showed a charge at that moment. It doesn't prove stable model identity, future availability, or support. Before production, you still need cost samples from real-sized requests, rate-limit and retry behavior, data retention terms, and a support contact.

Cloud and gateway credentials are separate contracts

If your company already runs on a cloud provider, you may prefer its IAM, procurement, regional controls, and combined billing. In that case the cloud account issues the credential and the cloud provider bills usage. Anthropic's Claude Platform on AWS documentation describes that setup. It doesn't add money to Claude Console, and cloud credentials and Console keys aren't interchangeable. The Max, Ultra, and Team monthly API credit doesn't cover Claude on Amazon Bedrock, Google Cloud Vertex AI, or Microsoft Foundry.

A third-party API gateway is another separate contract. It may offer one OpenAI-compatible endpoint for many models. But the gateway owns the balance, the token, the base URL, the logs, the data terms, and the support. Its token isn't an Anthropic key, and its credit isn't Anthropic credit. Before funding one, check its current model list, live prices, whether failed requests are billed, data terms, and who handles support. If you can't confirm any of those, don't use it.

Use Anthropic directly or a supported cloud when you need an Anthropic-issued credential, full Console controls, first-party support, or a specific procurement or data-residency contract.

The purchase rule that prevents most mistakes

Use Claude Console when you can sign up directly and want an Anthropic-issued key. Start with the new-user credit or your Max, Ultra, or Team monthly credit, then buy only what a few weeks of real usage needs. Use a supported cloud when cloud IAM, procurement, or regional control is the real requirement.

Never pay for a key you can't create, audit, rotate, and revoke yourself. The key opens the door; the account, the usage records, the endpoint, and the off switch are what make API access worth paying for.

Claude API key questions: students, Pro, Claude Code, and credits

Do you have to pay to get a Claude API key?

No. Creating a key in Claude Console is free. Calls made with it spend credits, and new users get a small free test credit first. After that, you buy prepaid credits, or use a Max, Ultra, or Team plan's monthly API credit.

Is an Anthropic API key free for students?

Anthropic's pricing FAQ lists no student API discount. Students get the same small new-user credit as anyone else. Researchers can apply to the External Researcher Access Program or AI for Science for API credits. For Claude chat discounts, see Claude Student Discount 2026: Official Answer, Real Routes, and Fake Code Checks.

Can I use my Claude Pro subscription as an API key?

No. Pro covers Claude chat and Claude Code on your Claude login, and it includes no API credit. Max, Ultra, and Team plans include a monthly API credit, but you still create a key in a linked Console organization to spend it.

Can you use the Claude Code API for free?

Not on an ongoing basis. Claude Code runs on a paid Claude plan login, such as Pro, Max, or Ultra, or on an API key that spends Console credits. The new-user credit can cover a short test with a key. The Max, Ultra, and Team monthly API credit doesn't pay for interactive Claude Code.

I already have a Claude subscription. How do I get my API key?

The same way as everyone else, because a subscription login doesn't come with a key. Sign in to Claude Console at platform.claude.com, open Settings → API keys, click Create key, and copy it once. On Pro, that key spends the new-user test credit and then prepaid credits you buy in Settings → Billing. On Max, Ultra, or Team, first claim the monthly API credit into one linked Console organization, as Anthropic's help article on monthly API credits describes; a key created in that organization spends the credit before any prepaid balance. Nothing you change in the Claude app alters what the key can spend.

What happens when my Claude API credits run out?

API calls stop until you add credits, unless auto-reload is on. Bought credits also expire one year after purchase, and Max, Ultra, and Team monthly credits expire at the end of each billing cycle.