Processors| Processor | What it does | Rate (per 1,000 pages) |
|---|
| Convert — fast / balanced | Document to markdown or HTML | 4 |
| Convert — accurate | Highest-fidelity conversion | 10 |
| Segment — page level | Detect boundaries between combined documents at the page level. Returns the page ranges and titles only — no OCR text | 0.5 |
| Segment — block level | Boundaries can fall mid-page at specific content blocks, driven by an optional custom prompt ("start a new segment at each invoice"). Returns the parsed markdown, so it runs Convert too | 4.5 |
| Extraction — turbo / fast | Structured fields, single pass. Turbo returns JSON only; fast adds per-field citations and runs on Convert | 6 |
| Extraction — balanced | Structured fields, multi-pass with per-field verification | 15 + fees* |
| Extraction — accurate | Structured fields, multi-pass with per-field verification, strongest model | 20 + fees* |
| Custom processor | Your own pipeline | 20 |
| Agent processor | Document agent workflows — the successor to custom processors, for higher-accuracy and more involved document work | 10‡ |
| Eval | Score output quality against a rubric | 2 |
| Form fill | Populate a form's fields | 10‡ |
| Create document | Generate a document | 6 |
Add-ons| Add-on | What it does | Rate (per 1,000 pages) |
|---|
| Track changes | Diff revisions in a Word document | 6 |
| Chart understanding (add-on) | Adds chart parsing to a run | +3 |
| Infographic (add-on) | Adds infographic parsing to a run | +4 |
| Word bounding boxes (add-on) | A box and confidence score for every word | +3 |
| Cross-page merging (beta) | Stitch content split across a page break back together: tables (long or wide), paragraphs broken mid-sentence, and lists | Variable** |
| Word bounding boxes | Identify bounding boxes at the word level with confidence scores so you can efficiently route OCR issues for human review and audit mistakes in your pipeline | +3 |
| Table cell bounding boxes (includes word bboxes) | Identify granular bounding boxes for each table cell (instead of the whole table). Also includes word level bounding boxes. | +6 |
| List bounding boxes (includes word bboxes) | Identify granular bounding boxes for each item in a list or list group (instead of the entire list as one block). Also includes word level bounding boxes. | +6 |
| Word, table cell, and list bounding boxes | Get all 3 granular bounding boxes at the word, table cell, and list level. | +9 |
* Around 5% of Balanced and Accurate extractions bill above our fixed per-page rates. That happens when the extraction agent has to
work through a document — long documents, and schemas that pull many repeated rows, are the usual cases; the rate covers the rest, and
you are never charged both the rate and the full compute. When it does apply it is typically a dollar or two, more for a long or
unusually dense document (Accurate runs a stronger model, so it both includes and can bill more). Schemas over 750 fields aren't
supported.
** Cross-page merging is in beta. It has no per-page rate — it bills only a variable compute surcharge reflecting the merge work your
document actually needs, which is typically a few cents per document.
‡ Form filling and agent processors run an agent over your document, so their cost follows the work it needs rather than its page
count. Almost every document bills this rate and nothing more; an unusually demanding one bills what the work actually cost instead.
The rate is a floor, not something a surcharge adds to, so you are never charged twice for the same run.
Spreadsheets bill by cells, not pages — 2,500 cells per page capped at $0.60 per sheet (simple) or 500 cells per page (intelligent),
chosen automatically. Processing in the EU adds 25% to the summed total. A single discount can also apply — 25% for opting into data
retention, or 33% under the startup program — but the two discounts can't both apply.
