Limits and billing
Size limits are checked when you submit, and a request over one is rejected with
413 and a code naming the limit. A request over a rate or concurrency limit is
rejected with 429 rate_limited and a Retry-After header giving the seconds to
wait.
One limit cannot be checked in advance: how much one job can summarize. A job that is too
large fails with documents_too_long, and one that runs out of time with
processing_timeout. Send fewer documents per job.
If your workload needs higher limits, contact support@emtelligent.com.
Billing
Jobs are billed in pages of text, to the organization that owns the key:
- one page per 2,048 bytes (UTF-8) of text in — the text of every document plus the prompt — rounded up;
- plus one page per 2,048 bytes of summary out, rounded up.
For example, 6.1 KiB in and 2.5 KiB out is 4 + 2 = 6 pages. Each result shows its
figure as stats.billed_pages.
What is billed is the text summarized, however it arrived — including the structured record
from fhir_bundles. The rest of what you send in nlp_results, PDFs in prep_documents
lines, skipped documents and other metadata are not counted.
Failed jobs are not billed. See Billing for credit and
invoices.

