Plans

The Plans page compares tiers and changes your subscription. Everything about what a plan costs is on Billing; this page is about what a plan gives you.

What a plan sets

A plan is not one number. It is several ceilings that move together, and knowing which one you are pressed against is the difference between a useful upgrade and an expensive one.

Credits

The monthly allowance, reset at the start of each billing cycle. Most endpoints cost one credit per call; heavier ones cost more, and each reference page states its own.

Credits do not roll over, and there is no overage — a spent balance means calls are refused, not billed. See billing.

Three details about how credits are counted are worth knowing before you size a plan:

Only successful calls count. A request rejected for bad input, one refused for a missing key, or one that fails on our side does not spend a credit. You are not charged for your own bugs, and a retry after a 500 costs the same as the call would have.

Cost is per call, not per row. Something that returns forty results for one query costs what its card says once. Where it gets expensive is calling the same source in a loop.

The balance is a live number you can read. Every response carries x-api-remaining-credits, so your own code always knows where it stands without opening this page. Analytics has the aggregate view.

Allowances reset at the start of each billing cycle — the anniversary of your subscription, not the first of the month. A plan started on the 12th resets on the 12th.

Rate limit and concurrency

Two different ceilings that both produce a 429, and they are the ones people most often mistake for a credit problem.

The rate limit is requests per minute. Concurrency is requests in flight at the same instant — usually what an unbounded Promise.all hits first, well before the per-minute figure. Rate limits has the backoff shape and a pool sized to your plan.

Sub-keys and team members

Sub-keys are scoped credentials for an environment, a client or a job. Team members are people who can hold their own keys and see usage without sharing a login. Both are the reason to upgrade when the problem is organisational rather than volume — one shared key passed around a team is a security posture, and not a good one.

Platform features

JSON bins and mock endpoints are counted per account. Embedded forms collect submissions straight into your APIs. GraphQL is available on every plan; the number is how much of it you get.

Other limits

Max response caps the size of a single response body — relevant on endpoints that return documents, rendered images or large result sets. AI tokens is the daily allowance for the AI Assistant, reset at midnight UTC.

Working out which ceiling you have hit

Start from what the platform told you, not from the plan page:

Upgrading for credits when the real constraint is concurrency buys you nothing, and it is a common enough mistake to be worth the extra thirty seconds of diagnosis. Analytics shows which one your traffic is actually pressed against.

Changing plan

Selecting a plan opens a confirmation that states what changes: your new credit allowance, your new rate limit, and the timing.

Upgrades take effect immediately, prorated for the unused part of the current period. The new credits and limits are live on your next request — there is nothing to redeploy, and your key does not change.

Downgrades take effect at the end of the current period. You keep what you have paid for until it runs out. If you are downgrading because usage fell, that is a month of headroom rather than a cliff.

Nothing about a plan change touches your key, your sub-keys, your usage history or your integration. Only the ceilings move. That is worth saying plainly because it is the question people ask before upgrading: there is no migration, no re-issued credential and no downtime, and a downgrade does not delete the sub-keys or team members you created above your new limit — they stop being editable, not existing.

The one thing a plan change does affect is premium fields. Endpoints omit fields your plan does not include rather than zeroing them, so an upgrade makes new keys appear in responses your code is already parsing, and a downgrade makes them disappear. Check for presence, not for a sentinel value.

Monthly or annual

The plans page has a Monthly / Yearly toggle, and every paid plan can be billed either way. Annual plans are the same plan — same credits, same rate limit, same features — charged once a year at a lower effective monthly rate, which the page shows as a struck-through monthly price next to the annual one.

Two practical differences worth weighing:

Annual is a year's commitment, so it suits a workload you already understand. If you are still learning what your usage looks like, a month or two on monthly billing tells you which plan you actually need, and switching to annual afterwards costs nothing.

Switching between the two is a plan change and follows the same rules as any other: to a higher effective price it applies immediately and is prorated; to a lower one it applies at the end of the period.

Annual plans do not appear as separate cards — the toggle changes what the existing cards mean, so there is one decision to make rather than eight.

Free trials

Paid plans can be started as a 7-day free trial with a trial credit allowance rather than the full monthly one:

No charge is made until the trial ends, and cancelling before then stops it entirely.

Eligibility depends on your subscription history: the trial is offered to a free account that has never had a payment record on it. An account that has subscribed before — even to a plan it later cancelled — is offered the plan directly instead. If you are building something that needs to know this in advance, the account summary reports isTrialEligible and trialCredits without spending a credit.

The trial is the plan's features at a smaller allowance, which makes it a real test of everything except volume: rate limit, concurrency, sub-keys and the premium fields all behave as they will after it converts. Point a staging environment at it for a week and you learn more than any amount of reading here.

Choosing a plan

Free is for building and evaluating. It is a real plan with real limits, not a time-boxed demo — small production workloads run on it indefinitely.

Starter is the first tier for something with users. The jump in rate limit matters as much as the credits.

Pro adds the organisational features: sub-keys, team members, and the scoping controls that let you stop handing the primary key to everything.

Mega adds IP allow-listing, per-key rate limits, audit history with 90-day retention, and the highest ceilings. It is where compliance requirements usually land you.

If none of that decides it, size on the shape of your traffic rather than the total. Steady background work — a nightly enrichment job, a queue drained at a fixed rate — is a credits question, and the cheapest plan whose allowance covers the month is the right one. User-facing traffic is a concurrency question, because it arrives in bursts that have to be served now; there the plan you want is the one whose concurrency ceiling your peak fits under, even if you never approach its credit allowance.

The pricing page carries current offers and the full feature matrix.

Next

Billing handles the subscription itself. Rate limits explains the ceilings above in terms of the responses you get when you reach them.