AIStall
EditionBy The AI Stall editorsOctober 28, 2026

The Monthly Cap on a New API Account

The Monthly Cap on a New API Account

A pay-as-you-go AI account is not unlimited. Each of the three large platforms caps what a new organisation can spend in a month, and the cap is published.

This note records those ceilings as the vendor documentation showed them on October 27, 2026. Every figure below was read from the rendered page source, not from a summary.

ChatGPT Business's public homepage, captured automatically on October 10, 2026. Not a logged-in account; the page may have changed since.
ChatGPT Business's public homepage, captured automatically on October 10, 2026. Not a logged-in account; the page may have changed since.

The reason to care is simple. A cap you did not know about stops your traffic mid-month, and two of the three platforms return the same HTTP error for a cap as for a rate limit.

Two vendors stop at the same number

OpenAI publishes four usage tiers with a monthly usage limit attached to each. Free is $100 a month. Build is $500, Launch is $5,000, and Grow is $200,000.

Claude Team's public homepage, captured automatically on October 10, 2026. Not a logged-in account; the page may have changed since.
Claude Team's public homepage, captured automatically on October 10, 2026. Not a logged-in account; the page may have changed since.

The qualification for each tier is a cumulative credit purchase, not a monthly spend. Build needs $5 in total credit purchases, Launch needs $100, and Grow needs $500.

That is a low bar for a large ceiling. A one-off $500 purchase moves an organisation from a $100 monthly limit to a $200,000 one, per the rate limits guide read October 27, 2026.

Anthropic publishes three named tiers with monthly spend caps: Start at $500, Build at $1,000, and Scale at $200,000. Organisations on the Custom tier have no monthly cap.

So both vendors stop their self-serve ladder at exactly $200,000 a month. Above that, both route you to an account team rather than a slider.

The difference is how you climb. Anthropic places organisations on a tier automatically, based on usage history and account standing. It does not publish a dollar threshold for each step.

Google caps the ten minutes, not just the month

The Gemini API documentation lists a billing tier cap per usage tier. Tier 1 is $250, Tier 2 is $2,000, and Tier 3 is written as a range: $20,000 to $100,000 and above.

A published range rather than a figure is unusual. It means the top self-serve tier is negotiated in practice, even though it appears on a documentation table.

Google also enforces something the other two do not publish: a spend-based rate limit measured over a rolling ten-minute window.

Read as a rate, $200 per ten minutes works out to $28,800 a day if sustained. That is the editorial arithmetic, not a figure Google prints.

The practical effect is different from a monthly cap. A burst of expensive requests can trip the ten-minute limit on a Tier 2 account that is nowhere near its $2,000 monthly ceiling.

Google's tier qualifications are the most explicit of the three. Tier 2 requires $100 paid plus three days from the first successful payment. Tier 3 requires $1,000 paid plus thirty days.

What the account does when it hits the ceiling

Anthropic documents the behaviour in detail. Once an organisation reaches its tier spend cap, API usage pauses until 00:00 UTC on the first day of the next month.

The response is HTTP 429 with error type rate_limit_error, the same type returned for an ordinary rate limit. There is no retry-after header, so automatic SDK retries keep failing.

The documentation gives a way to tell the two apart. On the Messages API the field error.details.error_code reads enforced_spend_limit_reached for a spend cap.

OpenAI's configurable hard spend limit also returns a 429 on affected requests. Its spend alert does the opposite: it sends a notification and lets traffic continue.

Google returns 429 RESOURCE_EXHAUSTED when a spend-based rate limit is hit, and the documentation suggests waiting and retrying or reducing the rate of expensive requests.

The cap you set yourself behaves differently

Both OpenAI and Anthropic let an organisation set a limit below the tier ceiling. On Anthropic, that self-set limit produces a different error from the tier cap.

Requests return HTTP 400 with error type invalid_request_error, and the message begins with a line about reaching your specified API usage limits. Raising the limit restores access immediately.

This matters for anyone writing retry logic. A 400 is not retryable and a 429 usually is, so the two caps need different handling in the same codebase.

The pattern is the same shape as the billing surprises in the add-on that is not in the seat price: the number exists, it is just filed somewhere a buyer does not look.

A new account may start below the published floor

One line in the Anthropic documentation deserves more attention than its placement suggests. New organisations may start in an Evaluation tier.

Limits in that tier sit below the standard limits shown on the page, while account history is established. They rise automatically as usage builds.

So the published Start tier numbers are not a guarantee for day one. A team planning a launch against those figures should check the console before it depends on them.

Google says something comparable in the other direction. Meeting the stated qualification criteria is generally sufficient for approval, but an upgrade request may in rare cases be denied after review.

What to check before the first invoice

Bottom line

The headline numbers are easy to summarise. OpenAI runs $100, $500, $5,000 and $200,000 a month. Anthropic runs $500, $1,000 and $200,000. Google runs $250, $2,000 and a range.

What differs is the mechanism. One vendor sells the next tier for a $5 purchase, one grants it on usage history, and one adds a ten-minute spend throttle on top of the monthly cap.

For token rates behind this spend, see the cached-input price note and the fifty percent batch discount.

For dated price changes on the same models, see the introductory price note. For the tiers that publish no price at all, see the trial that needs a sales call.

Teams buying seats rather than tokens should start with the ChatGPT Business review, the Claude Team review and the shortlist of AI assistant plans.

What this brief prints and what it cuts is in issue zero.

Sources