AWS Textract Pricing 2026: Official Rates Per 1,000 Pages

Last updated July 25, 2026. Every rate below was re-verified on that date against the official AWS pricing pages (US West, Oregon), including the newer Amazon Bedrock Data Automation rates.

AWS Textract pricing is pay-as-you-go by the page, and the rate depends on which API you call. In the US West (Oregon) region the published rates run $1.50 per 1,000 pages for Detect Document Text (plain OCR), $15 per 1,000 pages for Tables, $50 per 1,000 pages for Forms, $70 per 1,000 pages for Forms, Tables, and Queries together, about $10 per 1,000 pages for Analyze Expense, and about $25 per 1,000 pages for Analyze ID. There is no monthly subscription and no seat license. New AWS accounts get a three month free tier that covers 1,000 pages a month of plain OCR and only 100 pages a month of the Analyze Document features. This guide breaks down the 2026 rates per API, the volume discounts, the free tier limits, and the costs the headline per-page number leaves out.

Two things this guide does not cover, because they only became visible when we pulled the numbers straight from Amazon's machine-readable price list: AWS quietly discounts every multi-feature call by exactly $0.01 a page, and one US region charges 1.4 times the rest. Both are laid out with the full sixteen-meter table on our AWS Textract pricing rate card.

AWS Textract pricing per 1,000 pages (2026)

Textract does not have one price. It has a price per feature, and the gap between the cheapest and the most expensive feature is roughly 45x. Here are the published US West (Oregon) rates, with the discounted rate that kicks in after the first million pages in a month.

Two things about this table matter more than the numbers themselves. First, rates vary by AWS region, so confirm yours on the current AWS Textract pricing page before you build a budget on them. Second, the features stack. If you call Analyze Document asking for Forms and Tables on the same page, you are billed the combined rate for that page, not the cheaper OCR rate.

How is AWS Textract priced?

Textract bills per page processed, per feature requested, with no minimum commitment and no monthly platform fee. You are charged when a page passes through an API, and the rate is set by which API you called and which features you asked for in that call. Volume discounts apply automatically once you cross one million pages in a calendar month for most features.

A page is a page whether it is a clean digital PDF or a crooked phone photo of a receipt. Textract does not charge more for hard documents and it does not charge less for easy ones. What changes the bill is how many features you ask for. Teams that discover Textract is expensive are almost always calling Analyze Document with Forms turned on for every page, including the pages that only needed plain text.

Is there a free tier for AWS Textract?

Yes, but it is time limited and small. New AWS accounts get three months of free usage: 1,000 pages a month of Detect Document Text, 100 pages a month of the Analyze Document features (Forms, Tables, Queries, Signatures, Layout), 100 pages a month of Analyze Expense and Analyze ID, and 2,000 pages a month of Analyze Lending. After three months the free tier ends completely and every page bills at the standard rate.

One hundred free pages a month of Forms extraction is enough to run a proof of concept on a handful of documents. It is not enough to test whether Textract handles the twenty layouts your vendors actually send. Plan to spend real money during evaluation, or test on a platform that does not meter your pilot.

How much does AWS Textract cost at volume?

Run the arithmetic on your own document mix rather than on the headline OCR rate, because the headline rate is the one you will almost never pay. Three worked examples at 50,000 pages a month:

• Plain text only: 50,000 pages of Detect Document Text at $1.50 per 1,000 is $75 a month. Cheap, and you get unstructured text you still have to parse.
• Invoices through Analyze Expense: 50,000 pages at $10 per 1,000 is $500 a month, and you get named fields like vendor, date, and total.
• Mixed forms through Analyze Document: 50,000 pages at $70 per 1,000 for Forms, Tables, and Queries is $3,500 a month.

The million-page threshold is where the discount lands, and most US mid-market teams never reach it. If you process 50,000 or 200,000 pages a month, you pay the first-tier rate on every page. Budget accordingly, and do not assume the volume discount will rescue a spreadsheet that only works at scale.

Where the AWS Textract volume tier starts, compared to Azure and Google

This is the comparison that changes a procurement decision, and a list price table hides it completely. AWS Textract Detect Document Text, Azure AI Document Intelligence Read and Google Enterprise Document OCR all list at exactly $1.50 per 1,000 pages. Their volume tiers do not start in the same place.

Google's own pricing page states its first tier as "1 - 5,000,000 pages/month" at $1.50 and its second as "5,000,001+" at $0.60, verified August 2026. So anywhere between one and five million pages a month, AWS Textract has already halved and Google has not. At two million pages a month that works out to roughly $2,100 on Textract against roughly $3,000 on Google for identical basic OCR. Comparing the two on list price at that volume gives you exactly the wrong answer.

Worth knowing in the other direction too: this is specifically Google's basic OCR meter. Google Form Parser and Custom Extractor both tier at 1,000,001 pages like everyone else. And Google Layout Parser is published as "Single Tier: 1 - unlimited" at $10 per 1,000, so it never discounts at any volume.

The AWS exception to watch: most Textract meters discount above a million pages, but Queries does not. It stays at $15 per 1,000 pages at every volume. Analyze ID works differently again, dropping from $25 to $10 after the first 100,000 pages rather than at the million mark. If you build a pipeline on Queries and model your growth on the Detect Document Text curve, your forecast will drift the moment volume climbs. We keep the full cross vendor picture, including the Azure commitment tiers that are not published on the marketing pages, on our high volume OCR API pricing reference.

Why do Forms and Tables cost so much more than plain OCR?

Because they are different jobs. Detect Document Text recognizes characters and tells you where they sit on the page. Forms extraction has to work out that the characters "Invoice Date" are a label, that "03/14/2026" is the value that belongs to it, and that the two are related even though they sit forty pixels apart in a two column layout. Table extraction has to rebuild a grid that exists only visually. That inference is the expensive part, and it is the part you would otherwise write yourself.

This is worth understanding before you optimize your bill. The common cost mistake is routing every page through the most expensive call. The fix is classification: work out what each page is first, then send plain pages to cheap OCR and only send structured pages to Forms and Tables. Textract will not do that routing for you. You build it.

What does the AWS Textract per-page price not include?

The per-page rate buys you a JSON response. It does not buy you a working process. Costs that show up after the invoice does:

• Engineering time. Textract is an API. Someone has to stand up the AWS account, configure IAM roles, write the upload path, handle the asynchronous job flow for multi-page PDFs, map Textract's blocks and relationships into your field names, and keep it running.
• Classification. Textract will read whatever page you hand it. Deciding which document type that page is, so you know which fields to expect and which API to call, is your code.
• Human review. Textract returns a confidence score per field. Deciding what to do with a 71% confident total, building the screen where a person corrects it, and tracking who changed what is all yours to build. Amazon A2I exists for this and bills separately.
• Surrounding AWS services. S3 storage for the documents, Lambda or containers for the orchestration, data transfer, and CloudWatch logs all appear on the same bill under different line items.
• Validation and export. Checking that line items sum to the invoice total, that a date is a real date, and that the result lands in your ERP is application code.

None of this is a criticism of Textract. It is a well built, accurate, extremely scalable API and it is priced like one. It is simply not a product, and comparing its per-page rate to the price of a finished product compares two different things.

AWS Textract vs a ready-to-use extraction product on cost

Here is the honest version of the comparison. If all you need is text off a page and you already have engineers inside AWS, Textract at $1.50 per 1,000 pages is very hard to beat on raw cost, and you should probably use it. Nothing in this article changes that.

The picture changes when you need structured fields. Analyze Document with Forms, Tables, and Queries costs about $70 per 1,000 pages, and it still hands back JSON that your team has to classify, validate, review, and export. DocuOCR's published plans work out to roughly $14 to $20 per 1,000 pages depending on volume, and that price includes the classification step, the extraction, the validation rules, the human review screen, the dashboard, and the export, with no AWS account, no IAM policy, and no pipeline to maintain.

Pick on fit, not on the sticker. A team of engineers processing ten million pages of plain text should use Textract. An accounting or operations team that needs validated fields out of 30,000 mixed pages a month, this quarter, will spend less in total on a product that already includes the parts Textract leaves for you to write.

How do you estimate your AWS Textract bill?

Take a real month of documents, not a guess. Count the pages, not the files, because a 12 page loan package is 12 billable pages. Sort the pages by what you actually need from each one: text only, tables, form fields, or invoice fields. Multiply each bucket by its rate from the table above. Then add roughly 10 to 20 percent for the S3, Lambda, and data transfer line items that sit around the API, and add the fully loaded cost of the engineer who will build and own the pipeline. That last number is usually larger than the Textract line, and it recurs every year.

If most of your volume is invoices, price the Analyze Expense path specifically rather than assuming Forms, because the difference between $10 and $50 per 1,000 pages compounds fast. If you are doing this to get invoice line items into a spreadsheet at the end, be honest about whether a raw API is the shortest path to that spreadsheet.

Does AWS Textract charge for pages it reads badly?

Yes. Textract bills for pages it processes, and the confidence score it returns does not change the rate. A page it reads at 62% confidence costs exactly the same as a page it reads at 99% confidence. That matters because low-confidence pages are the expensive ones twice over: you pay Textract for the read, and then you pay a person to check it.

This is the cost line that never appears in a pricing comparison and always appears in a real budget. If 8% of your pages need a human to look at them, and a person handles 60 pages an hour, then 50,000 pages a month buys you roughly 67 hours of review labor. At $25 an hour that is about $1,675 a month, which dwarfs the $500 Analyze Expense bill sitting next to it. Any tool that reduces review volume, or that includes the review screen instead of making you build one, is competing on that number and not on the per-page rate.

What is the official AWS Textract Detect Document Text price per 1,000 pages?

Detect Document Text is $1.50 per 1,000 pages in US West (Oregon), dropping to $0.60 per 1,000 pages for volume above 1 million pages a month. That is the plain OCR path: it returns words, lines and their positions, with no key-value pairs and no tables. It matches Azure's Read model and Google's Enterprise Document OCR to the cent, so on plain text the three clouds are genuinely interchangeable on price.

The full official AWS Textract rate card (August 2026)

Re-verified against the AWS Textract pricing page on 14 August 2026. The figures AWS publishes there are quoted for the US West (Oregon) region, and they are stated per page, so the per-1,000 columns below are those figures multiplied by a thousand. The API operation column gives the identifier you actually call, because the display name and the operation name are not the same string.

What is the difference between DetectDocumentText and AnalyzeDocument pricing?

DetectDocumentText is $0.0015 per page, or $1.50 per 1,000. AnalyzeDocument starts at $0.015 per page with TABLES and rises to $0.05 with FORMS, so it costs 10 to 33 times as much. The gap is not a markup on the same work: DetectDocumentText reports words and lines, while AnalyzeDocument runs layout analysis on top of the recognition.

The practical consequence is that you cannot size a structured extraction project from the $1.50 figure, which is the number everybody quotes. If you need a table, you are on the $15.00 meter. If you need key-value pairs, you are on the $50.00 meter. You also pay for each FeatureType you request in the same call, so asking for FORMS and TABLES together bills both.

One detail worth knowing because it saves money: LAYOUT is included at no extra charge when you request it alongside TABLES, so if you are already paying for table structure you can have paragraph, heading, header and footer classification for nothing. There is no equivalent on the FORMS meter.

What neither operation gives you is a spreadsheet. Textract returns tables as Block objects and no operation emits CSV or XLSX, which we walk through in does AWS Textract output CSV. The cross-vendor view of which APIs do return CSV is on PDF to CSV.

Queries is the one line with no volume discount at all: $15 per 1,000 pages whether you send a thousand pages or ten million. Analyze ID is the opposite, with the steepest step-down of any Textract feature, from $25 to $10 per 1,000 pages after just 100,000 pages.

What is the AWS Textract free tier, exactly?

The Textract free tier runs for three months from your first request, not indefinitely, and the allowances differ by API. New AWS customers get 1,000 pages a month of Detect Document Text, 100 pages a month of Analyze Document (Forms and Tables), 100 pages a month of Analyze Expense, 100 pages a month of Analyze ID, and 2,000 pages a month of Analyze Lending. After three months every page bills at the standard rate. That 100-page monthly allowance on the structured APIs is small enough that a serious evaluation will exhaust it in an afternoon, so budget for a few dollars of paid testing.

Does AWS Textract offer a batch discount?

No. Textract's asynchronous operations exist because multi-page PDFs cannot go through the synchronous path, not to save you money, and AWS publishes a single rate per feature regardless of how the request arrives. Neither Azure nor Google discounts batch either. The only vendors that do are the LLM providers, Mistral and Gemini, both at 50% off. If someone has told you that queuing Textract work overnight will cut the bill, it will not.

Textract has no custom model and no classifier

This is the structural gap in Textract's price list, and it costs more than any per-page rate. AWS does not offer custom model training for document extraction, and it does not offer a document classifier. Azure charges $30 per 1,000 pages to run a custom extraction model you trained, and $3 per 1,000 pages to classify a mixed batch. Google charges $30 and $5. On AWS you cannot buy either, so a non-standard layout means Queries at $15 per 1,000 with hand-written prompts, or Forms at $50, plus whatever routing logic you build yourself. Compare the rate cards on the jobs you actually need, not on the OCR line where everyone matches.

Is Amazon Bedrock Data Automation cheaper than Textract?

For plain text, no, and it is not close. AWS now sells a second document extraction service, Amazon Bedrock Data Automation, and because it is newer and generative most people assume it is both the successor to Textract and the cheaper option. Neither is true. Textract is not deprecated, all of its meters are live, and AWS has announced no end of support. On price, Bedrock Data Automation is dramatically more expensive for the job most people actually run.

All rates per 1,000 pages, verified from the AWS pricing pages on July 13, 2026.

Bedrock Data Automation wins exactly one band: the full structured record where you would otherwise chain Forms, Tables and Queries together for $70 and then stitch three output shapes into one. Getting that back as a single schema you defined, for $40, is a genuinely better deal. Outside that band it loses, and it loses worst on the highest-volume job there is, which is simply reading the text.

Two further things the rate card does not advertise. Bedrock Data Automation charges $0.0005 per field per page above 30 fields in a blueprint, so a maxed-out 100-field blueprint costs $0.075 per page, which is $75 per 1,000 pages, more than Textract's most expensive bundle. And it has no volume discount on any meter, so at one million pages a month plain text extraction costs $1,500 on Textract and $10,000 on Bedrock Data Automation. The gap widens with scale rather than closing.

The practical conclusion is that this is not a migration decision, it is a routing decision. Keep Textract as the cheap default across the whole corpus and call Bedrock Data Automation only on the documents that genuinely need a structured record. Full detail on the Bedrock Data Automation pricing reference and the Bedrock Data Automation vs Textract decision guide, and the four ways to cut the bill are in how to reduce AWS document extraction costs.

The short answer on AWS Textract pricing

Textract is priced per page and per feature: about $1.50 per 1,000 pages for plain OCR, about $15 for Tables, about $50 for Forms, about $70 for Forms with Tables and Queries, about $10 for invoices through Analyze Expense, and about $25 for identity documents. Volume discounts start above one million pages a month. The free tier lasts three months and covers 100 pages a month of the features most people want to test. It is an excellent API and a genuinely cheap way to read text at scale, as long as you have budgeted for the pipeline, the classification, and the human review that surround it.

For the full cross-vendor picture, see our OCR API pricing comparison, which puts these rates side by side with Azure and Google. For the single-vendor breakdowns, read our Azure Document Intelligence pricing guide and our Google Document AI pricing guide. If you would rather skip the pipeline entirely, DocuOCR is a ready-to-use Amazon Textract alternative that returns validated fields instead of JSON blocks.

Price is only half the picture. Before you build, check the AWS Textract limits too, because the one-page synchronous cap and the 3,000-page asynchronous ceiling shape your integration as much as the per-page rate does. See how they compare with Azure and Google on the OCR API limits comparison.

For every vendor's rate on one normalized unit, see our OCR pricing per 1,000 pages reference. At enterprise volume, what it costs to OCR 1 million pages applies each discount tier, and does batch OCR cost less covers the discount that does not exist. For Textract's page and file-size ceilings, see AWS Textract limits.

Does handwriting cost extra on Textract?

No. Handwriting is read by the same DetectDocumentText operation as printed text and bills at the same $1.50 per 1,000 pages, with no handwriting add-on to enable. The constraint is language, not price. Amazon's FAQ states that "Handwriting, Invoices and Receipts, Identity documents and Queries processing are in English only", while printed text extraction covers English, German, French, Spanish, Italian and Portuguese. So turning to handwriting narrows Textract from six languages to one at no change in cost. We compare all three providers on this in our handwriting OCR reference.