How Claude API credits work, and what happens when they run out

How Claude API prepaid credits work

Most teams in the Claude Console pay through prepaid usage credits. Anthropic charges your balance down as you make requests, while organizations with an invoicing arrangement set up through the Sales team are billed monthly in arrears instead. Purchased credits cover Claude API access, the Console playground, and Claude Code. Note that if you run Claude Platform on AWS, billing is handled through AWS Marketplace rather than Anthropic credit purchases.

You're billed only for successful API calls and completed tasks, and failed requests aren't charged. But there's a catch that trips developers up. If your client disconnects or times out in the middle of a request that was on track to succeed, that request is still charged. Anthropic finished the generation on their backend, so those tokens deduct from your balance even if your server dropped the connection.

What happens when credits run out

If you run out of credits, you can no longer call the API or use the playground until you add more. Your requests don't queue or wait; they fail immediately.

Beyond running your balance to zero, you have to watch out for spend limits. Spend limits set a maximum monthly cost for API usage, whereas rate limits restrict request volume over a defined period.

When an organization reaches its tier's monthly spend cap, API usage pauses until 00:00 UTC on the first day of the next month unless you request a higher limit sooner. Requests return an HTTP 429 status code. The error type is rate_limit_error, but the response has no retry-after header, so automatic retries will fail until next month. On the Messages API, error.details.error_code is enforced_spend_limit_reached. That's how you tell a hard spend cap apart from standard rate throttling.

What happens if you set a custom spend limit instead? User-configured limits return an HTTP 400 with error type invalid_request_error. The message starts with "You have reached your specified API usage limits" (or workspace limits if applied to a specific workspace) and states when access resumes. Also, limits on the Claude Code workspace are checked separately; Claude Code requests exceeding that workspace limit can receive an HTTP 429 that carries a retry-after header.

Setting up and managing auto-reload

To prevent downtime when your balance runs dry, you can set up auto-reload to purchase credits automatically when your balance falls below a set threshold.

Here is how to configure it:

1. Log in to the Console with an Admin or Billing role.
2. Go to the Billing page at platform.claude.com/settings/billing.
3. In the Credit balance section, click "Set up auto-reload" (or "Manage auto-reload" if already on).
4. In "When credit balance reaches", enter the balance that triggers a purchase.
5. In "Bring credit balance back up to", enter the target reload balance.
6. Click "Turn on", or "Update" if changing existing settings.

Watch your balance before saving. If your balance is already at or below the minimum you set, the button reads "Turn on and purchase" or "Update and purchase", and you're charged right away to reach the target, plus taxes.

To disable it, click "Manage auto-reload", click "Turn off", and confirm by clicking "Turn off" again.

Expiration dates, bundled plan credits, and invoices

Purchased Claude API credits expire one year from the purchase date, and the expiration date can't be extended. All credit purchases are non-refundable. Expired credits appear in your Invoice history on the Billing page.

If you subscribe to Claude Max, Ultra, or Team plans, those plans include monthly API credits claimed by linking a Console organization to your plan:

• Monthly credits are spent before purchased credits.
• Unused monthly credits expire at the end of each billing cycle.
• Monthly API credits can't be used in Claude Code or the Claude apps, including for extra usage.

Organizations using prepaid billing receive a credit purchase receipt for every charge, including auto-reloads. Anthropic automatically emails the invoice to your billing email with the subject line "Your receipt from Anthropic."

Admins can download past invoices on the Billing page under Invoice history. Invoices below $0.50 aren't charged on their own; they roll forward as an applied balance on your next invoice. Issued invoices cannot be modified, so billing detail changes apply only to future invoices. For bank transfers, payment must match the invoice amount exactly or it won't apply, taking around five business days to process once received.

Tier spend caps and custom limits

Every account operates under a default tier spend cap. You can set custom spend limits below your tier cap on the Billing page by clicking "Adjust limit" or "Set limit", though spend limits cannot be set on the default workspace.

Standard usage tiers enforce the following monthly spend caps:

Organizations with an invoicing arrangement set up through the Anthropic Sales team bypass prepaid credits entirely. Anthropic aggregates usage across API calls, Console usage, and other services at standard pay-as-you-go pricing, sending a Stripe invoice at the end of each calendar month.

Catching runaway spend with Costbrake

Auto-reload prevents outages, but a rogue loop can drain your balance and re-trigger charges repeatedly.

Costbrake monitors spend across 23 providers - including Anthropic, OpenAI, AWS, and Cloudflare. Checks run hourly on the Free plan ($0) and every 5 minutes on Pro ($12 a month) and Team ($39 a month), dropping to 1-minute checks during a spike. A spike alert triggers when an hour costs more than 3x your 7-day baseline and at least $5 above normal. You also get alerts for budget thresholds at 50%, 80%, and 100%, 72-hour burn forecasts, and new charges over $1.

On Pro and Team, Costbrake provides opt-in brakes. For Anthropic, Costbrake connects to the Usage and Cost report APIs using an Admin key. If a spike hits, you can disable an individual API key or disable all active keys in a workspace (up to 60 keys per brake) with one click and confirmation. It records the prior state so you can undo it immediately once you patch your code. Brakes never fire on their own, and Costbrake doesn't delete resources.

Tell your AI coding agent "Set up Costbrake for me so I get alerted when my cloud and AI bills spike. Follow the instructions at https://costbrake.com/agent" or check Costbrake pricing to get started.