How much is the OpenAI API costing me this month?

Last updated: 26 September 2026 · Deutsch

Sign in at platform.openai.com, open the Usage page and pick the Cost tab: it shows what your organisation has spent this month, in US dollars, and it adds up to your invoice. Outside the browser, the Costs API returns the same figure — but only to an admin key. Your ChatGPT plan is not part of it; it is billed on its own.

Where the number is

Sign in at platform.openai.com and open the Usage page. Pick the organisation and the month; the Cost tab gives the breakdown that matches the invoiced amount. You need to be an organisation owner or hold the usage dashboard permission to see it.

The page has its own project selector, separate from the project picked elsewhere on the platform — check it before you trust a low number. Since August 2026 you can also filter and group by API key. Dates run in UTC, so in Switzerland the new month starts at 01:00 or 02:00 local time, not at midnight.

What is left of your credit is not on the Usage page but on the billing page of the organisation settings.

Prepaid credit or monthly invoice

OpenAI bills the API in one of two ways. With prepaid billing you buy credit first, and each request draws it down. With monthly billing you get an invoice after the calendar month ends; credit bought in advance is set against it.

When the credit runs out, requests fail with the error code credit_balance_exhausted. The stop is not instant: usage in the gap can push the balance below zero, and OpenAI takes that from your next purchase.
Auto recharge buys credit when the balance drops below a threshold you set. Its optional monthly limit caps those automatic purchases — not what you spend.
Credit expires one year after purchase and is not refunded.

Three numbers are easy to mix up. Usage counts requests and tokens, not dollars. Costs are the dollars billed, day by day — the figure that ties back to the invoice. The credit balance is what is left of prepaid credit. Top up mid-month and balance and costs tell different stories: for this month's spend, look at costs.

Can I cap the spend?

Yes. Since July 2026 OpenAI offers two controls, for the whole organisation or for one project. A spend alert sends a notice and lets traffic run on. A hard spend limit makes requests fail with a 429 error once tracked spend reaches the amount.

Organisation: Settings → Organization limits → Spend → Edit spend limit. Enter the monthly spend limit; turn on “Enforce a hard limit” if requests should stop.
Project: Project settings → Limits → Spend → Edit spend limit, with the same fields.
Which one hit shows in the error code: organization_spend_limit_exceeded or project_spend_limit_exceeded.

Four things to know. You need the right to manage the organisation or project settings. Enforcement is not instant, so spend can end up slightly above the limit. A reached limit holds until you raise or remove it, or the next month starts. And apart from your own limits, OpenAI sets a usage limit for your organisation by usage tier; hitting that one returns organization_usage_limit_exceeded. OpenAI's advice: put an alert below the hard limit, so you hear about it before traffic stops.

Why the figure lags

OpenAI counts costs in whole days in UTC and says the usage data may be delayed. It names no fixed delay. So a request from a few minutes ago may not show yet, and today's figure keeps growing until the day closes.

For a month that is over, the Cost tab is the figure that matches the invoice.

Asking the API yourself

The same figures come as JSON from the Costs API. It needs an admin key: only organisation owners can create one, under Admin keys in the organisation settings (platform.openai.com/settings/organization/admin-keys). It begins with sk-admin-. A normal project key gets a 401 — and an admin key in turn cannot call models.

curl -G https://api.openai.com/v1/organization/costs \
  -d start_time=1788220800 \
  -d bucket_width=1d \
  -d limit=31 \
  -H "Authorization: Bearer $OPENAI_ADMIN_KEY"

start_time is in Unix seconds; 1788220800 is 1 September 2026, 00:00 UTC.
Buckets are always one day wide. Without limit you get only seven days; the most is 180.
Add up amount.value over all results in data, and you have the month so far.
group_by splits by project_id, line_item or api_key_id. If has_more is true, fetch the next page with page set to next_page.

Token counts for the same period come from /v1/organization/usage/completions. Two things the Costs API does not give you: your prepaid balance, for which the Admin API reference lists no endpoint at all, and your limits. Those have their own Admin API endpoints, to read and to change spend limits and alerts.

Does a ChatGPT subscription cover API costs?

No. ChatGPT and the API platform have separate billing. A Plus or Pro fee does not count as API credit, and API usage does not show up on your ChatGPT bill. Each has its own billing settings and history.

So if the question is “what does the API cost me”, the ChatGPT side does not enter into it — and the other way round.

The same at Anthropic

The Claude Console works much the same way. Its Cost page shows spend by model and month, spend limits and alerts are set per workspace, and the cost report comes from /v1/organizations/cost_report with an Admin API key. As with OpenAI, a Claude subscription and API usage are billed apart.

What AI Cockpit does with this

AI Cockpit shows this month's cost of the OpenAI and Anthropic APIs in the Mac menu bar, on iPhone and iPad, and on Apple Watch. It calls the same cost endpoints named above, with the admin key you enter, and it only ever reads: every request is a plain GET. On the Mac, the OpenAI card shows:

today, the current month and the total, in dollars as OpenAI reports them;
cost per project and per model, plus this month's input and output tokens;
a note when today runs at 1.8 times or more your median of past days.

It refreshes every ten minutes by default. It does not read OpenAI's new spend limits yet; instead you can enter your own monthly limit, and the bar turns orange at 75 per cent and red at 90. The balance it can only work out: the amount of your last top-up, minus what was spent since that date.

Which services it reads, why each needs the key it does, and where the data stops: Keeping track of AI API costs. Short answers on admin keys and empty cards are on the FAQ page.