How to cap OpenAI, Anthropic and Gemini API spending

API keys are the cheapest way to use strong AI models, until a loop, a long document or a forgotten script runs all night. This guide covers what the providers give you, where those tools stop, and how a cap that is checked before each call closes the gap.

Step 1: use what your provider offers

Each provider's console has some way to watch spending. Look for these, and set them today:

• A monthly limit or budget on the account or the project, where the provider offers one.
• Email alerts at a share of that budget, so you hear about a spike while it happens.
• One key per app or project, so one runaway tool cannot spend the money of the others, and you can revoke it alone.

Read your provider's own page for what each setting does. Some are hard limits; some are only alerts that tell you after the money is spent. The names and rules change, so check them rather than trusting a blog post, this one included.

Step 2: know where those limits stop

Provider limits share three gaps:

• They count after the fact. Usage is added up as calls finish, so a burst can pass the line before it is enforced.
• They are per provider. A budget at OpenAI knows nothing about your Anthropic or Gemini spending.
• They are per account, not per task. They cannot say "this one job may cost at most this much".

Step 3: cap each task before it runs

Cortex66 adds the missing piece: a cap that is checked before a call leaves, across every provider you use.

• The cap comes first. The most a task can cost is reserved before anything is sent. If it does not fit, it does not run.
• An unknown price never runs. A model whose price cannot be worked out is not called.
• Caps nest. Workspace, person, project, run and step each can have a cap, and the tightest line wins.

Before a task runs you see a receipt with the model, the reason, and the most it can cost. The full rule set is in One cap for every AI. Your keys stay yours: Cortex66 uses the keys you already pay for first and tells you when a lighter model would have done the same job.

Read more

• Bring your own key: what it costs and what it protects
• ChatGPT Plus or an API key: which costs you less?

FAQ

Can I set a hard spending limit on the OpenAI API?

Check the limits and budget settings in your OpenAI console; what is available depends on your account. Add alerts either way.

Why did my AI API bill go over my budget?

Most limits are counted after calls finish, so a burst of calls can pass the line before the limit takes effect.

How do I limit spending across several AI providers?

Provider budgets are separate. A cap checked before each call, like Cortex66's, counts every provider against one line.

What happens in Cortex66 when a task would pass the cap?

It does not run. The receipt shows the most it could cost, and you choose a lighter model or a higher cap.