Glossary · APIs and AI agents
What are API credits?
The short answer
API credits are prepaid units that a service deducts each time a program calls its API. The provider decides how many credits a call or a result costs, so the price of a job depends on how many lookups it runs, how many results come back and, with some providers, whether a lookup that finds nothing is charged too.
Real data
What three real calls counted
Three calls to a data API that meters in records rather than credits, with the count each response returned. One record is one item returned; a lookup that finds nothing counts nothing.
Returned by the BlitzAPI API on
/v2/jobs/searchA job search that matched 252 postings and returned 3
- Postings that match
- 252 job postings
- Postings returned
- 3 job postings (max_results set to 3)
- Counted
- 3 records
The search: US job postings from the last 30 days with Revenue Operations Manager in the title.
/v2/enrichment/domain-to-linkedinA domain lookup that finds the company
- Domain
- stripe.com
- Found
- Yes, the Stripe company page
- Counted
- 1 record
/v2/enrichment/domain-to-linkedinA domain lookup that finds nothing
- Domain
- no-company-here-blitz-glossary.com (made up for the test)
- Found
- No
- Counted
- 0 records
Each count is the fair_usage.records_used field of the response, the same value as its x-records-used header.
How do API credits work?
With API credits, you buy a balance up front, or receive one with a plan. Each call to the API draws it down by an amount the provider sets: one credit per request, one per result, or more for a costly endpoint such as a phone number. When the balance is empty, calls stop or are billed as overage until you buy more.
The model is popular because one balance can pay for very different endpoints. The price of a credit, though, says little on its own. What a job will cost depends on what each call consumes.
What does one call cost in API credits?
Two rules decide what one call costs in API credits, and both are in the provider’s documentation rather than on its pricing page.
- What is counted. A request, a result, or a field. A search that matches 252 job postings and returns 3 costs one unit under a per-request meter and three under a per-result meter, like the first call above.
- Whether a miss is counted. A lookup that finds nothing, like the third call above, costs nothing on some APIs and a full credit on others. On a cold list, misses can be a large part of the calls.
Show the answer key
- 252That is how many postings matched the filters. A per-result meter counts what comes back, not what matched.
- 3Three results came back, so three units are counted. The 249 others were never returned.
- 1One is the count under a per-request meter, where every call costs the same whatever it returns.
- 0A call that returns results always uses something. Zero is what a lookup that finds nothing can cost, on APIs that do not charge misses.
API credits, usage-based pricing and flat plans: what is the difference?
API credits, usage-based pricing and flat plans differ in when you pay and in what sets the bill:
| API credits | Usage-based pricing | Flat plan | |
|---|---|---|---|
| How you pay | A prepaid balance | An invoice after the period | A fixed monthly price |
| What sets the bill | The credits each call draws | The units you consumed | The plan: a request rate and a monthly volume |
| What to watch | Expiry, the cost of misses | Spikes from a runaway script | The volume the plan covers |
Usage-based pricing is the same meter billed afterwards. A flat plan moves the question from “how much will this job cost?” to “does this plan cover the job?”, which is easier to forecast. Requests per second, the other meter of a data API, are covered on the API rate limit page.
Why are API credits hard to forecast?
API credits are hard to forecast because the cost of a job depends on numbers you only know after running it: how many lookups come back empty, how many results each search returns, how many calls a retry adds. A team that budgets from the list size alone will be off by the share of misses, in one direction or the other.
The fix is a small test batch on a sample of the real list. Count what came back, what was charged, and what was usable, then scale. The same habit helps whatever the meter, including the agent purchases described in agentic commerce.
How does BlitzAPI count usage instead of API credits?
BlitzAPI counts usage in records, not credits. One record is one item the API returns to you: one company, one person, one email address, one phone number or one job posting. A call that finds nothing counts nothing. Each plan covers a number of records a month, its fair use policy, and a number of requests per second on each endpoint: the pricing page lists them.
Every metered response returns the records it used, so the cost of a job can be read call by call, and GET /v2/account/key-info returns the records left before a large run, at no cost (API reference).
Sources and method
- Written by
- Antoine Blitz, Founder, BlitzAPI
- Published
- API calls run on
- BlitzAPI docs, Fair use policy
- BlitzAPI docs, Errors and response headers (x-records-used)
- BlitzAPI API reference, Search Jobs
- BlitzAPI API reference, Domain to LinkedIn URL
The calls ran on October 28, 2026 through the BlitzAPI MCP server, which calls the same endpoints; they are shown here as the equivalent REST requests. Only the per-call count is shown: the balance of the account used for this glossary is not a public plan.
Credits are described here as other providers use them. BlitzAPI does not sell credits: its plans count records against a fair use policy.