Your documents never leave your bucket. We read from your prefix and write the results back beside the source. Nothing is retained after a job finishes, and credentials are dropped the moment it ends — which is the part a hosted OCR API cannot offer you at any price.
You buy credits and spend them on whichever output you need. Every plan includes a monthly allowance that resets on the first of the month.
| Output | Credits per page | What you get |
|---|---|---|
layout | 2 | Markdown, real <table> HTML with rowspan and colspan intact,
and every region typed with its pixel box. |
fields | 5 | Key/value pairs by name — alternative labels, plain-language descriptions, and per-document-type rules. |
Bulk text-only extraction is available in the API at a lower credit cost; ask if you have an archive to run.
2,000 credits a month.
1,000 document pages, or 400 pages with field extraction.
Field extraction included.
20,000 credits a month.
10,000 document pages, or 4,000 pages with field extraction.
Field extraction included.
40,000 credits a month.
20,000 document pages, or 8,000 pages with field extraction.
Field extraction included.
Need more part-way through a month? Add credits to your balance; they are spent at the same per-page rates and carry over. Only pages we successfully read are charged — failed pages cost nothing.
Independently measured on OmniDocBench with the official evaluator, our table structure scores 0.912 TEDS — level with the published state of the art. See for yourself on your own document before you pay us anything.