How to Cap OpenAI, Claude & Cloud API Bills

What is the difference between an alert and a hard cap?

An alert sends you a message when spending crosses a threshold. A cap refuses further requests. Most cloud billing tools are alerts only. AI providers offer real caps, but they are not instant, so a small overshoot is normal.

Real people do get caught out. In May 2026 The Register reported that Google Cloud and AWS customers had been billed from about $3,000 to over $100,000 after exposed or abused API keys and automatically raised spending tiers, and that a user billed directly by Anthropic was protected by its spending limit. These were individual user reports, and a follow-up covered refunds.

How do you set a hard limit on OpenAI?

OpenAI separates spend alerts from hard spend limits. Alerts notify you and let traffic continue. A hard limit makes affected requests fail with a 429 error once monthly spend reaches it.

1. Create one OpenAI project per app so each has its own limit.
2. For the whole organization: open the Organization limits page, choose Spend, then Edit spend limit, enter a monthly amount and turn on Enforce a hard limit.
3. For one app: open that project's settings, go to Limits, then Spend, and repeat the same steps.
4. Handle the error in your app. A project limit returns project_spend_limit_exceeded and an organization limit returns organization_spend_limit_exceeded.

OpenAI notes that enforcement is not instantaneous, so recorded spend can slightly exceed the amount you set. Pick a limit below what you can afford.

How do you set a spend limit on Anthropic (Claude)?

In the Claude Console, open the Billing page and find Spend limits. Click Set limit (or Adjust limit) and enter an amount. Your limit cannot exceed your usage tier's monthly cap.

• Use a separate workspace for each app and set a workspace spend limit. Spend limits cannot be set on the default workspace.
• When a limit you set is reached, requests return an HTTP 400 invalid_request_error that says when access resumes. Retrying will not help until you raise the limit or the month rolls over.
• Organization-wide limits always apply, even if workspace limits add up to more.

Spend limits are separate from rate limits, which cap requests and tokens per minute and protect against bursts rather than monthly cost.

What can you do on AWS?

Use two tools together. AWS Budgets lets you set a monthly cost budget with alerts on actual and forecasted spend, sent by email or Amazon SNS, and optional budget actions, such as applying an IAM policy that denies provisioning more resources. AWS Cost Anomaly Detection uses machine learning to flag unusual spend.

• Budgets data updates up to three times a day, and AWS warns that charges can pass your threshold before the notification arrives.
• Cost Anomaly Detection relies on Cost Explorer data that can lag by up to 24 hours, and a new monitor can take 24 hours to start detecting.
• AWS says Cost Anomaly Detection does not monitor most AWS Marketplace products, with third-party foundation models on Amazon Bedrock as the exception. For other Marketplace charges, use a Budgets cost budget filtered by billing entity. Check how your own models are billed.

Neither tool is a hard cap on its own. Keep AI keys out of public code, scope IAM permissions tightly and delete unused keys.

What can you do on Google Cloud?

A standard Google Cloud budget does not cap usage. Google's documentation says setting one does not automatically cap spending. Alerts default to 50%, 90% and 100% of actual spend and go to billing admins by email.

• Add forecast-based thresholds and extra recipients, such as a Cloud Monitoring channel.
• For automation, send budget notifications to a Pub/Sub topic and build a function that responds, for example by disabling billing on a project. Google presents this as something you build, and notices can arrive hours late.
• Google also documents spend cap budgets, but only for supported services. Check whether yours is eligible.
• Restrict every API key to the specific APIs and referrers or IP addresses it needs.

What about Vercel and other hosts?

On Vercel's Pro plan and Enterprise Flexible Commitment plans, Spend Management lets an owner or billing user set an On-Demand Budget. Notifications go out at 50%, 75% and 100%, and you can add a webhook.

• Setting the amount does not stop usage by itself. Turn on Pause Production Deployments to pause all projects when the amount is reached. Visitors then see a 503 error until you resume each project manually.
• Vercel checks spend every few minutes, so pausing can lag. Set the amount below your true maximum.
• Pausing does not stop AI Gateway key usage, which still counts toward spend.

Other hosts differ. Look for a budget, a spend limit and an option to pause or throttle.

What should you build into the app itself?

• Keep provider keys on the server only. A key in the browser is a key anyone can use.
• Require login before any AI call and rate-limit per user.
• Set max_tokens (or the provider's equivalent) on every request and cap conversation length.
• Cache repeated answers and use a cheaper model for simple tasks.
• Log token usage per user so one account cannot silently use most of your budget.

Our production checklist and security guide cover the first two in more detail. If you want us to look at your setup, the free audit call includes spend caps, caching and model routing.

Sources

Checked 27 Oct 2026. Provider settings and limits change, so confirm against the linked documentation.

• OpenAI: spend limits and alerts
• Anthropic: rate limits and spend limits
• AWS: managing costs with AWS Budgets
• AWS: Cost Anomaly Detection
• Google Cloud: budgets and budget alerts
• Vercel: Spend Management
• The Register (18 May 2026): surprise AI bills for AWS and Google Cloud users
• The Register (15 May 2026): Google reimburses affected users

Keep going

• Free AI App Security & Cost Audit call
• AI-built app development and rescue
• Vibe-Coded App Production Checklist: 12 Steps Before You Launch
• How to Secure a Vibe-Coded App Before Launch