AWS Textract pricing, decoded: what your OCR bill will really be

Amazon Textract looks cheap until the first real bill. A team wires up document extraction, tests it on a few hundred invoices, sees a couple of dollars on the AWS console, and ships. Then volume arrives - and the bill is not a little higher, it is 20 or 40 times higher than the back-of-envelope guess. Nothing broke. They simply picked the wrong API for the job and never read the price list past the first line. Textract pricing is not complicated, but it is counter-intuitive: the same page can cost $1.50 or $70 per thousand depending on one choice you make in code. Here is exactly what each call costs in 2026, the worked math on a real invoice workload, and when Google or open-source OCR is the smarter spend.

The one choice that sets your Textract bill: which API you call

People say "Textract pricing" as if there is one number. There is not. Textract is a family of APIs, and each one is priced for a different job. Your monthly cost is decided almost entirely by which of these you call - not by clever optimisation later:

• Detect Document Text - raw OCR. Gives you the words and their positions, nothing structured. The cheapest call.
• Analyze Document - structured extraction, priced per feature you switch on: Forms (key-value pairs), Tables, Queries (ask "what is the invoice total?"), Signatures, and Layout. Turn on more features, pay for each.
• Analyze Expense - a purpose-built model for invoices and receipts. Returns vendor, totals, dates, and line items without you configuring anything.
• Analyze ID - passports, licences, and identity documents.
• Analyze Lending - mortgage and loan document packages.

The trap is that Analyze Document with Forms and Tables can read an invoice, so teams use it for everything. It is the most expensive way to do the most common job. The bill that follows is not Textract being pricey - it is the wrong tool billed at premium rates.

The actual 2026 price list (US region, per 1,000 pages)

These are the standard on-demand prices in the US region. The first number is the rate for the first 1 million pages per month; the second is the discounted rate above 1 million (Analyze ID steps down after the first 100,000). Prices vary slightly by AWS region, so confirm yours in the console - but the ratios between APIs are what matter, and they hold everywhere.

Two things to read off this table. First, the features on Analyze Document add up: Forms is $50 and Tables is $15, so an invoice run with both is $65 per 1,000 - and $70 once you add Queries. Second, Layout is included at no extra charge when you already use Forms, Tables, or Queries, and Signatures is free when bundled with another feature, so you rarely pay the standalone $3.50. The expensive line items are Forms and the combined Analyze Document call; everything cheap lives at the top of the table.

A worked example: the same 20,000 invoices, three ways

Say you process 20,000 single-page supplier invoices a month - a realistic volume for a growing D2C brand or a mid-market finance team. Here is the monthly Textract cost three ways, for the exact same pages:

• Detect Document Text (you only need the raw text, and you parse fields yourself): 20 x $1.50 = $30/month.
• Analyze Expense (the invoice-specific model returns vendor, total, tax, and line items for you): 20 x $10 = $200/month.
• Analyze Document with Forms + Tables (the general-purpose call teams reach for by habit): 20 x $65 = $1,300/month.

Same 20,000 pages. $30 versus $1,300 - a 43x swing, and it climbs past 47x once you add Queries. The team paying $1,300 is not getting better invoice data; Analyze Expense is actually more accurate on invoices because it was trained for them. They are simply paying a premium to use a general tool for a specialised job. This single decision - matching the API to the document - is worth more than every other Textract cost tweak combined.

Textract vs Google Document AI vs Cloud Vision: when each wins

Textract is rarely the only option, and on price it is not always the winner. The honest comparison depends on what you are extracting:

• Plain OCR (just the text): it is effectively a tie. Textract Detect Document Text, Google Cloud Vision text detection, and Google Document AI Enterprise OCR all sit around $1.50 per 1,000 pages, with Cloud Vision giving you the first 1,000 units a month free. Pick on ecosystem, not price. We break the accuracy differences down in our Textract vs Google Document AI OCR comparison.
• Structured forms and tables: Google Document AI Form Parser is about $30 per 1,000 pages, roughly half of Textract Forms + Tables at $65. If your workload is generic key-value and table extraction at scale, Google is usually the cheaper structured option.
• Invoices and receipts specifically: Textract Analyze Expense at $10 per 1,000 is hard to beat on total cost for that one job - it is purpose-built, needs no configuration, and is far cheaper than routing invoices through general Forms extraction on either cloud.
• Just the raw pixels-to-text, no structure: Cloud Vision is the lightweight pick, especially for images rather than multi-page PDFs.

The pattern: Textract wins on invoices and on staying inside an AWS-native pipeline; Google Document AI wins on general structured forms at scale; Cloud Vision wins on cheap plain OCR of images. There is no single cheapest provider - only the cheapest provider for your document type. If you are already running the rest of your stack on AWS, our AI on AWS team usually keeps document extraction there and tunes the API choice rather than moving clouds for a marginal per-page saving.

The free tier, and how a proof-of-concept blows through it

Textract has a 3-month free tier for new AWS accounts, but it is deliberately lopsided, and that catches teams out:

• Detect Document Text: 1,000 pages/month free.
• Analyze Document (Forms, Tables, Queries): only 100 pages/month free, each.
• Analyze Expense and Analyze ID: 100 pages/month free.
• Signatures: 1,000 pages/month free; Analyze Lending: 2,000 pages/month.

So if you are prototyping plain OCR, you can test for weeks on the free tier. But the moment your proof-of-concept touches structured invoices or forms, 100 pages is an afternoon of testing - and page 101 is billed at the full rate. Plan your evaluation around the 100-page structured cap, not the friendlier 1,000-page text cap, or the "free" pilot quietly starts costing money mid-test.

When open-source OCR is the cheaper call

At high volume, or when you only need plain text, self-hosted open-source OCR (Tesseract and friends) can undercut $1.50 per 1,000 pages, because you pay only for the compute you run, not a per-page API fee. It is a genuine option - but the honest trade-offs decide it:

• It shines when volume is very high, you need plain text only, and you have the engineering capacity to run and tune it. Millions of clean, typed pages a month is where the per-page API fee stops being trivial and your own infra wins.
• It struggles the moment you need structured key-value or table extraction. Tesseract gives you text and coordinates, not "this is the invoice total" - you would build the layout and field logic Textract Analyze Expense hands you for $10 per 1,000. That engineering, plus its ongoing maintenance, is the real cost, and it usually dwarfs the API savings at SME volumes.

The break-even is less about page price and more about people: managed Textract trades a per-page fee for zero ops; open-source trades a low compute bill for a permanent engineering commitment. For most teams under a few hundred thousand structured pages a month, Analyze Expense or the right Analyze Document call is cheaper once you count the salaries. Above that, and for plain text, it is worth modelling both - which is the kind of build-vs-buy call our AWS compute cost breakdown and our AWS consulting team run before anyone commits.

Frequently asked questions

Why is my Textract bill so much higher than $1.50 per 1,000 pages?

Because you are almost certainly calling Analyze Document with Forms rather than Detect Document Text or Analyze Expense. The $1.50 rate is raw OCR only. The moment you switch on Forms ($50) and Tables ($15) you are on a different tier entirely, and the features stack. Check which API your code calls before you assume the pricing is wrong - it usually is not.

Is Textract cheaper than Google Document AI?

It depends on the document. For plain OCR they are a tie at around $1.50 per 1,000 pages. For general structured forms, Google Document AI Form Parser (about $30) is cheaper than Textract Forms + Tables (about $65). For invoices and receipts specifically, Textract Analyze Expense at $10 is usually the cheapest accurate option on either cloud. There is no universal winner - match the provider to the document type.

Does Textract charge per page or per document?

Per page. A 10-page PDF processed with Detect Document Text counts as 10 pages. This matters for multi-page invoices and contracts - a "cheap per document" mental model will underestimate the bill. Always estimate in pages per month, not documents.

Can I lower the cost without changing accuracy?

Yes, usually a lot. The biggest lever is using the right API - Analyze Expense instead of Analyze Document for invoices often cuts the bill 5-6x with better accuracy. After that: only turn on the Analyze Document features you actually parse, filter out blank or duplicate pages before you send them, and batch asynchronous jobs. None of those touch the quality of the extracted data.

The short version: Textract is not expensive, but it is easy to overpay by picking a general API for a specialised job. Estimate your volume in pages, match each document type to the cheapest accurate call, size your pilot around the 100-page structured free tier, and only reach for open-source when plain-text volume genuinely justifies the engineering. If you want that modelled against your real documents before you build, our AI development and AWS consulting teams do exactly that - and it is what the Amazon Textract overview and our invoice-processing walkthrough build on.