1,000 Invoice Pages: $1.50 to Read the Text, $10 to Extract the Fields

The quote changes when the required output changes

For 1,000 pages in AWS US West (Oregon), outside the free tier, Amazon Textract DetectDocumentText lists $0.0015 per page: $1.50 for text recognition. AnalyzeExpense lists $0.01 per page in the same region and volume band: $10 for invoice or receipt data extraction. These are two different APIs for two different jobs, not a hidden surcharge on one OCR request. If your application needs a vendor name, total, dates, or line items in structured fields, the $1.50 text-only quote does not describe that task. If plain searchable text is enough, the $10 expense API may be unnecessary.

Page price is only the first layer. Real invoice-processing cost also depends on document length, extra processing attempts, human review, downstream work, and how many submitted documents your business accepts. Use the planning calculator below with your own assumptions; it does not predict extraction quality.

Check the Amazon Textract pricing examples and the AnalyzeExpense response fields. These US list-price examples were checked on September 21, 2026. They exclude a qualifying free tier, negotiated discounts, taxes, and extra workflow services. Confirm the current rate before purchase.

First specify the output, then compare the meter

An OCR API may return words and their positions. A layout or document parsing API may preserve reading order, tables, and key-value relationships. A specialized invoice processor may return normalized fields such as supplier, invoice date, amount, tax, and line items. Those outputs overlap, but they are not interchangeable. The cheapest way to obtain text is not necessarily the cheapest way to produce a usable accounts-payable record.

AWS documents the expense field and line-item response; Google lists the Invoice Parser's fields; and Cohere describes Parse v5's output. Match those outputs against a sample of your own documents before treating any price as a procurement answer.

AWS: one region, one volume, two separate API quotes

AWS's published US West (Oregon) examples put DetectDocumentText at $0.0015 per page and AnalyzeExpense at $0.01 per page for volumes below one million pages per month. At 1,000 pages, that is $1.50 versus $10. Both rates are list-price examples outside the free tier. The comparison isolates the API choice; it does not say that every invoice pipeline must run both APIs or that every customer will pay those amounts.

Each row assumes all pages are processed by just the named API in Oregon, all are in the first million-page tier, and no free allowance applies. For example, 10,000 × $0.0015 = $15 for text, while 10,000 × $0.01 = $100 for AnalyzeExpense. These are alternative API-only scenarios, not a prediction of your total bill. AWS documents lower rates after one million pages; the simple multiplication above should not be extended beyond that tier.

New AWS customers may qualify for a three-month free tier. AWS lists up to 1,000 DetectDocumentText pages per month, but only up to 100 AnalyzeExpense pages per month during that eligible period. Eligibility and timing matter: neither allowance should be silently applied to an established paid account. If you use the website calculator for AWS text OCR, turn on its trial option only when your account qualifies. The calculator does not expose AnalyzeExpense and cannot apply its separate allowance to an invoice estimate.

Source: AWS Textract pricing, free tier, and examples. Review the selected region in your own AWS account before budgeting.

Google: the invoice meter is a count, not the OCR page rate

Google's Enterprise Document OCR price card lists the first 1,000 OCR pages at $0 and then $1.50 per 1,000 pages through its next published tier under the standard consumption model. That allowance applies to that OCR processor; it is not a blanket free allowance for specialized parsers. The Invoice Parser is listed at $0.10 per count under the standard consumption model, and Google defines one count as up to 10 pages in a document.

That distinction changes the math. One thousand separate, single-page invoices represent 1,000 Invoice Parser counts, or $100 at $0.10 per count under those assumptions. One hundred invoices of 10 pages each represent 100 counts, or $10. Both groups contain 1,000 pages, but they do not produce the same specialized-parser estimate. These are arithmetic illustrations, not claims that both document sets have the same business value, support conditions, or quality. Check request limits, supported document types, and your contracted consumption model before use.

Source: Google Document AI pricing and count definition. The AICostBudget calculator covers Enterprise Document OCR, not Google Invoice Parser; its page input cannot infer how many invoice-parser counts your documents create.

What the other calculator cards mean

The calculator compares five named products, not five interchangeable invoice processors. Alongside Amazon Textract DetectDocumentText and Google Enterprise Document OCR, it displays Azure Document Intelligence Read, Cohere Parse v5, and Mistral OCR 4.1. Azure Read is a text-reading product, while Cohere Parse and Mistral OCR produce richer document parsing output. Neither a capability label nor a low page price proves that an API returns the exact invoice fields your accounting system expects.

Cohere advertises Parse v5 at $1.50 per 1,000 pages through its API, and Mistral lists OCR 4.1 standard processing at $4 per 1,000 pages. Those are useful rate-card facts, not evidence that either product is a drop-in replacement for AnalyzeExpense or Invoice Parser. Azure's public pricing page describes separate Read and prebuilt invoice meters but may render the regional price as a placeholder; inspect the tool's cited price record and verify your own Azure region and SKU before relying on a number.

Official references: Cohere Parse, Mistral API pricing, and Azure Document Intelligence pricing.

Turn a page quote into a task budget

Start with a representative document set. Count pages and documents separately, record document type and region, and write down the fields the application must return. Then choose the smallest API output that meets the acceptance criteria. If text OCR is followed by a parser or language model, price that second call explicitly. If one specialized API already includes OCR in its output, do not assume a separate OCR call is required. Finally, estimate retries, storage, and human review as separate, measured or labeled assumptions.

monthly_workflow_budget = selected_API_charges
  + downstream_processing_charges
  + retry_charges
  + storage_charges
  + human_review_cost

Do not add a charge for a step your implementation does not call.

The formula is a checklist, not an estimate with invented values. A production integration may also face limits, unsupported files, compliance requirements, and exception handling. Test your real documents and keep failed or reviewed items in the denominator when you calculate cost per accepted invoice. A cheap page price cannot compensate for an output that requires expensive repair, but that repair cost must be measured rather than guessed.

What the provider OCR calculator can and cannot settle

AICostBudget accepts a monthly page count and estimates the five products named above using its currently verified pricing records. It displays the applied pricing tiers, applicable unconditional allowances, an optional AWS new-customer trial, AWS commercial region choices, verification dates, and links to official sources. You do not upload a sensitive document for this estimate; the calculation runs in your browser. A stale or incomplete price is excluded from its cheapest-price ranking.

Its ranking is a cost-only result for the products and assumptions shown. It cannot decide which API will meet an invoice-field schema, compare extraction accuracy, estimate AnalyzeExpense or Google's Invoice Parser, infer document counts from pages, or calculate a complete invoice automation bill. Use it to narrow a page-priced OCR budget, then create a separate worksheet for specialized extraction and operational steps.

A pre-purchase checklist for invoice OCR

• Define the accepted output: raw text, layout, normalized invoice fields, line items, or a combination.
• Measure monthly documents and pages separately; keep the page-count distribution, not just the average.
• Name the exact API and consumption model before copying a provider's headline price.
• Check whether the billable unit is a page, a document, a count covering several pages, or a workflow step.
• Check the selected region, volume tier, contract, and free-tier eligibility for that specific API.
• Confirm whether OCR is already included in a specialized processor's output before adding another call.
• Test real invoice formats and measure retries, review time, storage, and any later model calls.
• Check data residency, retention, file support, and request limits against your procurement requirements.

If the job truly ends at searchable text, a text OCR estimate may be the right starting point. If it ends at a posted invoice record, treat field extraction and validation as part of the definition of the job. The useful comparison is between implementations that deliver the same accepted output, with each billable step shown plainly.

FAQ