← Back to articles

OCR and document extraction

Compare OCR API costs by output and billing unit

Plain text OCR, layout parsing, annotated extraction and token-priced vision LLMs are not interchangeable. Define the required output, then estimate the charge for that task.

Published Nov 9, 2026Official sources checked Nov 9, 2026

Select the output

Three different document tasks

Text recognition

Convert scanned pages to text. Check the page rate and whether layout metadata is included.

Layout and forms

Extract sections, tables, or form fields. Managed document processors may price different features separately.

Annotations

Request bounding boxes or richer extraction. Check model capability and its applicable annotated-page rate.

Image-capable language models can instead bill for image inputs and generated tokens. Do not treat a per-page OCR rate as a token price or assume equal extraction quality. If you send PDFs directly to OpenAI, Claude or Gemini, use the PDF input cost guide to estimate their different billing units.

Official model examples

Match the exact serving route to the published page rate

The mistral-ocr-latest alias is not an independent provider offer. Do not infer its future version or rank these routes by price without matching capabilities.

Worked budget

10,000 pages at published OCR 4.1 rates

Standard OCR

10,000 × $0.004 = $40

Annotated OCR

10,000 × $0.005 = $50

Different volume?

Billable page count × verified per-page rate. Check any additional services separately.

Illustrative list-price totals, not an invoice. These exclude discounts, storage, preparation, taxes, retries and unverified cache or Batch eligibility. Do not add the annotated and standard rates without confirming the provider's billing rules.

How to compare across providers

  1. Output: decide whether text, layout, tables or annotations are required.
  2. Workload: count billable pages and any additional operations.
  3. Route: verify API ID, region, quotas, free tiers and the official price page.
  4. Total: compare matched functionality; unknown rates are not zero.

For adjacent services, Google Document AI lists different processor pricing for OCR, layout and extraction. It is not an interchangeable LLM route. Browse OCR catalog entries for source-backed model details. The token calculator is not a native page-priced OCR estimator.