// Decision guide, July 13, 2026

Bedrock Data Automation vs Textract: Cost, Features and Which to Use

Textract is not deprecated, and the newer service is not the cheaper one. Bedrock Data Automation costs 6.7 times more than Textract for plain text, and 20% less than Textract for a structured form. The right answer for most teams is to run both.

  • A 16-row capability matrix
  • Every rate verified from AWS
  • The architecture that uses both
  • Where each one genuinely wins
See the matrix Extract a document free
Upload a document, no signup

PDF, JPG, PNG, BMP, HEIC, TIFF

Upload a document to extract

Extract documents →

Neither AWS service will tell you which fields it got wrong. Drop a document in and see what a reviewed extraction looks like.

SOC 2 Type II
256-bit encryption
US data handling
Fields, not just text
6.7x
more expensive for plain text
20%
cheaper than Textract Forms
4x
more than Analyze Expense on invoices
0
end-of-support dates announced
// The short answer

Does Bedrock Data Automation replace Textract?

No. Both services are live, both have active billing meters, and AWS has announced no end of support for Textract. They are built for different jobs and priced accordingly. Textract is the cheap, precise, per-feature OCR API: $1.50 per 1,000 pages for plain text, $15.00 for targeted queries, $10.00 for invoices through a purpose-built model. Bedrock Data Automation is a generative service that takes a blueprint describing the fields you want and returns exactly that schema in one call, at $40.00 per 1,000 pages. It is worth that rate when the alternative is chaining Textract Forms, Tables and Queries together for $70.00 and mapping three output shapes yourself. It is a bad deal for everything else, and it is a particularly bad deal for plain text extraction, where it costs 6.7 times what Textract charges for a job Textract does better.

The one-line rule

Pay Bedrock Data Automation for a schema. Never pay it for OCR.

If you can describe what you want as a list of named fields and the layout keeps changing, it earns its rate. If you just want the words off the page, Textract does that for $1.50.

// Feature by feature

Bedrock Data Automation vs Textract, compared on 16 counts

Rates per 1,000 pages, verified from the AWS pricing pages on July 13, 2026. Textract wins more rows than it loses, which is not the result most people expect from a newer service.

Capability Bedrock Data Automation Amazon Textract Winner
Plain text extraction, per 1,000 pages $10.00 (standard output) $1.50 (Detect Document Text) Textract
Structured fields, per 1,000 pages $40.00 (custom output) $50.00 (Forms) Bedrock Data Automation
Forms, tables and queries together $40.00 (one call, one schema) $70.00 (bundled, three shapes) Bedrock Data Automation
Targeted questions of a document $40.00 (custom output) $15.00 (Queries) Textract
Invoices and receipts $40.00 (custom output) $10.00 (Analyze Expense) Textract
Identity documents $40.00 (custom output) $25.00 (Analyze ID) Textract
Output shape A schema you define in a blueprint Generic key and value pairs you map Bedrock Data Automation
Tolerance of layout drift High, a blueprint describes intent Low, you remap when layouts change Bedrock Data Automation
Charges more for a wider schema Yes, $0.0005 per field above 30 No, no field concept Textract
Volume discount None at any scale Drops to $0.60 above 1M pages Textract
Languages, documents Six, no CJK, no vertical text Wider, includes CJK Textract
Modalities Documents, images, audio, video Documents only Bedrock Data Automation
Max pages per document 3,000 with the splitter 3,000 async PDF Tie
Handwriting Yes, no surcharge Yes, no surcharge Tie
Bounding boxes and per-block confidence Less granular Yes, precise geometry Textract
Purpose-built document models No, you write blueprints Yes, expense, ID, lending Textract

Full meter-by-meter breakdown on Bedrock Data Automation pricing and AWS Textract pricing.

// The answer nobody gives you

Do not migrate. Run both, and route by document.

The framing of this comparison as a migration is the mistake. These two services sit at different price points because they do different amounts of work, and the cheapest correct pipeline uses each one where it is strong.

1

Textract across everything

Run Detect Document Text at $1.50 per 1,000 pages over the whole corpus. That gives you searchable text, a page count and enough signal to classify what each document is. At this rate you can afford to do it to every page you own.

2

Route by what you found

Most documents do not need a structured record. The ones that do are usually a minority: the invoices you are going to pay, the forms that feed a system, the contracts someone has to act on. Send only those onward.

3

Bedrock Data Automation on the subset

Call custom output at $40.00 per 1,000 pages only on the documents that earn it. You pay the expensive rate on the pages that justify it instead of on the entire archive.

What the split actually saves

Take 100,000 pages a month where 10,000 of them need a structured record. Run everything through Bedrock Data Automation custom output and you pay $4,000. Run the split instead: 100,000 pages of Textract text at $1.50 per 1,000 is $150, plus 10,000 pages of custom output at $40.00 per 1,000 is $400, for a total of $550. Same result, and you have kept the searchable text for the other 90,000 pages that you would not otherwise have had. The reason this works is that Bedrock Data Automation has no volume discount, so there is never a scale at which sending it everything starts to pay off.

// Straight trade-offs

What each service is actually good at

Bedrock Data Automation wins when

  • Layouts drift. A blueprint describes what a purchase order number is, not where it sits. When a supplier redesigns their form, the extraction keeps working. A Textract Forms integration usually needs remapping.
  • You want one schema, not three shapes. Forms, Tables and Queries return three different structures and cost $70.00 per 1,000 pages bundled. Custom output returns the object you asked for at $40.00.
  • The document is not a standard type. There is no Textract prebuilt model for a lease abstract or a lab report. There is a blueprint for anything you can describe.
  • You also have audio or video. Documents, images, audio and video through one API, which Textract does not attempt.

Textract wins when

  • You just need the text. $1.50 against $10.00 per 1,000 pages, and it drops to $0.60 above a million. This is the majority of real OCR work.
  • The document is a standard type. Analyze Expense reads invoices and receipts at $10.00 per 1,000 pages. Analyze ID reads licenses and passports. These are trained models, not prompts.
  • You need a few specific answers. Queries costs $15.00 per 1,000 pages. Asking a document five questions does not justify a $40.00 blueprint.
  • You need geometry. Precise bounding boxes and per-block confidence scores, which matter if you are building a review interface that highlights where each value came from.
  • Your documents are not in one of six languages. Bedrock Data Automation handles English, German, Spanish, French, Italian and Portuguese for documents, and no CJK at all.
// Frequently asked

Bedrock Data Automation vs Textract: FAQ

Does Bedrock Data Automation replace Amazon Textract?
No. Both services are live with active billing meters, and AWS has announced no end of support for Textract. They are built for different jobs. Textract is a cheap, precise, per-feature OCR API. Bedrock Data Automation is a generative service that returns a schema you defined. For plain text extraction Textract remains dramatically cheaper and should stay the default.
Which is cheaper, Bedrock Data Automation or Textract?
It depends entirely on the job. For plain text, Textract wins by a wide margin: $1.50 per 1,000 pages against $10.00, so 6.7 times cheaper. For a full form with a schema, Bedrock Data Automation wins: $40.00 per 1,000 pages against Textract Forms at $50.00, or against the Forms plus Tables plus Queries bundle at $70.00. For invoices, Textract Analyze Expense wins again at $10.00.
Should I migrate from Textract to Bedrock Data Automation?
Only for the parts of your pipeline where you are currently chaining several Textract calls to build a structured record. If you are calling Detect Document Text to get raw text, migrating to Bedrock Data Automation will multiply that line of your bill by about 6.7 and buy you nothing. The better pattern for most teams is not migration at all, it is keeping Textract as the cheap default and calling Bedrock Data Automation only on the documents that genuinely need a schema.
What can Bedrock Data Automation do that Textract cannot?
Three things. It returns output in a schema you define in plain language, rather than generic key and value pairs you have to map yourself. It tolerates layout drift, so a supplier changing their invoice template does not break your extraction. And it handles documents, images, audio and video through one API, where Textract is documents only.
What can Textract do that Bedrock Data Automation cannot?
It gets much cheaper at volume, dropping to $0.60 per 1,000 pages above a million where Bedrock Data Automation never discounts at all. It supports more languages, including Chinese, Japanese and Korean, which Bedrock Data Automation does not handle for documents. It has purpose-built models for invoices, IDs and lending documents. And it gives you precise bounding boxes and confidence scores per detected block.
Is Bedrock Data Automation more accurate than Textract?
Neither AWS nor any independent benchmark publishes a head-to-head accuracy number, so treat any confident claim with suspicion, including ours. What can be said structurally is that a generative model handles unfamiliar layouts better because a blueprint describes what you want rather than where it sits, while a purpose-built model like Analyze Expense is usually stronger on the specific document type it was trained for. Test both on your own documents before deciding.
What are the page limits for Bedrock Data Automation?
The synchronous API caps at 10 pages and 50 MB. The asynchronous API allows 20 pages from the console, and with the splitter enabled it handles up to 3,000 pages and 500 MB per document. The often-repeated 20-page limit is the console figure, not the ceiling. Textract asynchronous processing also handles up to 3,000 pages for a PDF, so on long documents they are effectively tied.
Does Bedrock Data Automation support handwriting?
Yes, and at no extra charge. It reads printed and handwritten characters at the same rate, and it handles any in-plane rotation, so a page scanned at an angle still reads. Text has to be at least 15 pixels tall, which is roughly 8 point type at 150 DPI. Textract also reads handwriting, so this is not a point of difference between them.
Can Bedrock Data Automation and Textract be used together?
Yes, and for most teams that is the right architecture. Run Textract Detect Document Text at $1.50 per 1,000 pages across everything to get text, search and triage. Then call Bedrock Data Automation custom output at $40.00 per 1,000 pages only on the subset of documents where you actually need a structured record. You pay the expensive rate on the pages that justify it, not on the whole corpus.
Does Bedrock Data Automation work with Chinese or Japanese documents?
No. Document support is limited to English, German, Spanish, French, Italian and Portuguese, and AWS states that vertical text alignment is not supported, which rules out traditional Japanese and Chinese layouts. If you process CJK documents this is a hard gate, and it is the single most overlooked limitation of the service.
// Keep comparing

Every other OCR rate, verified the same way

Bedrock Data Automation pricing

Every meter, the 30-field surcharge, and the per-1,000-pages math.

Read it

AWS Textract pricing

Full rate card, the 3-month free tier, and the models Textract does not have.

Read it

OCR pricing per 1,000 pages

Every vendor normalized onto one unit, including the per-document ones.

Read it

Azure Content Understanding vs Document Intelligence

The same decision on the Microsoft side, where the newer service is the cheaper one.

Read it

Amazon Textract alternative

What to use instead of Textract, and when a managed pipeline beats a raw API.

Read it

OCR API pricing comparison

The buyer pillar: which OCR API to actually pick, by workflow.

Read it

Both of them hand you JSON and walk away

Whichever AWS service you pick, you get an API response. You still write the classifier that decides which documents need a schema, the review queue where a person fixes the fields the model got wrong, the validation that checks the line items sum to the total, and the export into your ERP. That engineering is usually a bigger line in the budget than the difference between $1.50 and $40.00 per 1,000 pages.

DocuOCR runs at roughly $14 to $20 per 1,000 pages with all of it already built. Worth comparing before you scope a build.

Test it on your own document

A comparison table cannot tell you whether the fields come out right on the forms you actually process. Upload one and look.

Extract a document free

Last updated July 2026. Rates read from the official AWS pricing pages and re-verified before publication.

Z tej samej rodziny narzędzi