AWS Textract vs Apryse 2026
Amazon Textract is a cloud API that turns scanned and digital documents into text, tables, form fields, and query answers. Apryse is an embeddable document SDK whose Smart Data Extraction, OCR, and handwriting (ICR) modules do similar work inside your own infrastructure, including your own AWS account, and whose viewing, editing, conversion, and compliance capabilities go well beyond it. This page compares the two where they overlap and is explicit about where they don't. This page helps you decide.
Quick Comparison
Join the ranks of over 20,000 innovative start-ups, governments and Fortune 500 companies, including 85 of the Fortune 100, that trust Apryse.
Amazon Textract is a service you call. Apryse is a component you ship.
Textract is, for many developers, the dependable baseline for cloud OCR. It is inexpensive for raw text, its table extraction holds up well in third-party testing and it plugs into S3, Lambda, IAM, and the rest of AWS with no new vendor to onboard. AnalyzeExpense returns typed invoice fields with no setup at all. If your pipeline already lives in AWS and your documents are printed English, it is a strong default.
The difference is in what you are buying. Textract is a service: each synchronous call takes one page, anything longer goes through an S3-based asynchronous job, every feature you enable is metered on top of the last, recognition covers six Western European languages, and by default AWS may use what you send to improve its AI services unless your organization opts out. Apryse is a library: Smart Data Extraction, OCR, and ICR run in-process on your Windows or Linux servers, or on your own Lambda functions and ECS tasks, with no network round-trip, no per-page meter, no page cap beyond your hardware, and OCR across 80+ languages. The same SDK also contains the viewer, editor, converter, and redaction engine that a document product needs around extraction, none of which Textract provides.
Whether that matters depends on your use case. For an AWS-native pipeline at modest volume over English documents, a metered API is simpler. For a product you ship to customers, a regulated or disconnected environment, multilingual documents, or volumes where per-feature pricing stops being viable, a component you own behaves differently.
"The rendering performance, combined with how responsive the Apryse engineering team has been, has made a meaningful difference in our product quality."
Amanda Lansman, Sr Product Manager, Dropbox
Feature Chart: See How Apryse Compares
This chart covers only the capabilities both vendors offer. Capabilities unique to Apryse are listed in the next section.
OCR engine and languages
Default OCR Module (SDK 12.0 and later): deep-learning engine with models for English and multilingual models covering 80+ languages across Latin, Cyrillic, CJK, Devanagari, and Arabic scripts. Optional IRIS iDRS module (10 languages including Chinese, Japanese, Korean) for complex layouts; a leaner 6-language Alternative module for constrained hardware.
Detect Document Text and Analyze Document recognize printed text in six languages: English, French, German, Italian, Portuguese, and Spanish — unchanged since November 2020. Latin character set only; no vertical text; the detected language is not returned. Supports all in-plane rotations, superscripts and subscripts (June 2025 update). Materially narrower language coverage than Apryse.
Handwriting (ICR)
Handwriting ICR Module (released April 2026): recognizes handwritten text in PDFs and images on-premises, outputs a searchable text layer and JSON with page numbers and word coordinates. English only.
Handwriting is recognized within the same OCR calls, English only.
OCR output
Searchable PDF, or JSON/XML nested as pages, paragraphs, lines, and words with coordinates, font size, and orientation. No per-word confidence score in OCR output. External OCR JSON can be applied back onto a PDF.
A JSON list of Blocks (PAGE, LINE, WORD) each with a 0–100 confidence, a normalized bounding box and polygon, rotation angle, and page number. JSON only from the API; searchable PDF, CSV, Excel, Markdown, and HTML are produced by the open-source Textractor library or your own code.
Layout and document structure
Document Structure engine returns paragraphs, headings (H1–H6), lists, tables of contents, tables, and images as JSON with a bounding rectangle and page number per element; distinguishes section columns from table columns. Structured Output additionally converts PDF to reflowed DOCX, XLSX, PPTX, and HTML. Currently, no chunking API and no Markdown output.
Layout feature returns ten element types (title, section header, header, footer, page number, list, figure, table, key-value, text) in implied reading order, column by column on multi-column pages, each with bounding box and confidence. Markdown and HTML linearization for LLM pipelines comes from the Textractor library, not the API.
Table extraction
Tabular Data engine handles merged cells and multi-row headers, outputs table, row, and cell JSON with row and column spans, bounding rectangles, page numbers, and a group ID that links a table continued across pages, or exports directly to XLSX. Deep Learning Assist improves table recognition in the Document Structure engine. No per-cell confidence score.
Tables feature returns table, cell, and merged_cell blocks with row and column spans, column headers, table titles, footers, section titles, and summary cells, typed as structured or semi-structured; each with geometry and confidence. $15 per 1,000 pages. A table continued on the next page is returned as a separate table, whereas Apryse recognizes these.
Forms and key-value pairs
Form Field Detection finds text fields and checkboxes in scanned or static PDFs (radio buttons announced as coming) and can write fillable fields back into the PDF. Form Field Key-Value adds label-to-value matching. Generic Key-Value Extraction is template-free across document types, returning key, value, word-level rectangles, and a 0.0–1.0 confidence per pair. Also extracts CAD title-block metadata from engineering drawings.
Forms feature links key and value blocks and detects checkboxes and radio buttons as selection elements with a selected or not_selected status, each with geometry and confidence. $50 per 1,000 pages (the most expensive single feature) and the usual reason combined calls reach $65–$80 per 1,000 pages. Form models are not trainable.
Queries and custom extraction
Not available. No natural-language queries, custom schemas, user-defined fields, or model fine-tuning in the SDK. The documented pattern is to pass Smart Data Extraction JSON to the LLM of your choice for schema mapping. A separate .NET Template Extraction product supports user-defined template zones.
Queries answer natural-language questions per page (15 synchronous, 30 asynchronous; English only; cannot return whole tables, rows, or columns) with an answer, location, and confidence. Custom Queries adapters tune Queries on 5–2,500 labelled documents.
Signature detection
Form Field Detection identifies signature fields as part of form-field output with coordinates and confidence. The wider SDK applies and validates digital signatures (see next section).
Signatures feature returns the location and confidence of handwritten signatures, electronic signatures, and initials, optionally tied to a key-value pair or table cell. Free with any other feature; $3.50 per 1,000 pages standalone. Detection only, it does not sign or validate.
Classification and splitting
Page-level classification into 24 predefined categories (SDK 11.12 and later): invoices, receipts, IDs, passports, forms, letters, legal documents, financial reports, technical drawings, and more; each with a 0.0–1.0 confidence and a configurable threshold. No user-defined labels. Splitting is done with the SDK's page-manipulation APIs rather than a model.
No general-purpose classifier or splitter. The only classification is AnalyzeLending's fixed set of 35 mortgage document types. Blueprint-based classification and semantic splitting of files up to 3,000 pages exist in Amazon Bedrock Data Automation, a separate service.
Coordinates, page references, confidence
Typed JSON for every engine. Every element carries a rectangle and a 1-based page number. Confidence scores on form fields, key-value pairs, and classification results; table and structure elements carry coordinates but no confidence.
Every Block carries a 0–100 confidence, a normalized bounding box and polygon, and a page number (AnalyzeID fields excepted, which carry confidence but no geometry). Parity on coordinates; Textract additionally scores every word and cell.
Reproducibility and version control
Structure is recovered deterministically before models interpret it, and the SDK version is pinned by you: the same document on the same release produces the same headers, tables, and fields on every run.
Pretrained models rather than generative ones, and developers describe the output as stable for a given model. Model updates are applied by AWS service-side. Responses report an AnalyzeDocumentModelVersion, but no request parameter selects one. So output can change when AWS updates the model. Only Custom Queries adapters have selectable versions.
Page and size limits
No documented per-document cap in licensed use; long documents are processed whole and the licensed module has no page limit.
Synchronous operations accept one page and 10 MB; a multi-page PDF returns UnsupportedDocumentException. Multi-page PDFs and TIFFs must be uploaded to S3 and processed with the asynchronous Start and Get operations (500 MB, 3,000 pages), with completion signalled through SNS; results are retrievable for 7 days.
Pricing
Package license: Smart Data Extraction, OCR, IRIS, and Handwriting ICR are add-on modules to the Server SDK, licensed rather than metered per page or per operation; document volume is an input to the quote, not a runtime meter.
Per page, per feature. DetectDocumentText $1.50 per 1,000 pages ($0.60 above 1M a month). AnalyzeDocument: Tables $15, Queries $15, Forms $50; Forms and Tables $65; Forms, Tables, and Queries $70; Forms, Custom Queries, and Tables $80. Layout $4 but free with Forms, Tables, or Queries; Signatures $3.50 but free with any other feature. Custom Queries $25 (adapter training free, no free tier). AnalyzeExpense $10; AnalyzeID $25 ($10 above 100K); AnalyzeLending $70. Free tier for three months: 1,000 OCR pages and 100 Forms, Tables, or Queries pages a month. S3, SNS, and Lambda charges for the asynchronous workflow are extra.
Support and SLA
Maintenance and support Monday–Friday (Pacific) with a two-business-day response guarantee; enterprise tiers negotiated. No uptime SLA applies because the software runs in your environment.
Service credits when monthly uptime falls below 99.9% (10%), 99% (25%), or 95% (100%). Technical support through AWS Support plans (restructured December 2025): Basic free; Business Support+ from $29 a month or 9% of monthly AWS charges with 30-minute critical response; Enterprise from $5,000 a month or 10% of charges with 15-minute response.
Beyond the Overlap — The Rest of the Apryse SDK
Amazon Textract is an extraction API. Apryse is a document SDK in which extraction is one module. The capabilities below are part of the same Apryse license and codebase, and are not part of Textract at all. AWS addresses some adjacent needs with other services (Bedrock Data Automation, Amazon Comprehend, Amazon Q), but not within Textract.
Viewing and rendering
WebViewer: client-side WebAssembly rendering of 30+ formats (PDF, Office, CAD, images, email) with a source-available React UI; document content never leaves the browser. Native iOS and Android viewers.
Not part of Textract.
Annotation and collaboration
Annotation and collaboration 35+ annotation types, real-time multi-user collaboration, measurement tools, document comparison, XFDF round-trip.
Not part of Textract.
Forms, signatures, redaction
AcroForm and XFA forms, form creation and filling, electronic and digital signatures with timestamps and long-term validation, true redaction with pattern search, PDF sanitization.
Not part of Textract. (Textract detects where a signature is; it does not sign, validate, or redact.)
Editing
PDF text and image editing, page manipulation, native in-browser DOCX editing with track changes and comments, in-browser spreadsheet editing.
Not part of Textract.
Conversion
Office to PDF without Microsoft Office, PDF to DOCX/XLSX/PPTX/HTML (Structured Output), HTML to PDF, CAD to PDF (DWG, DXF, DGN, DWF; RVT in beta on Windows), email (EML, MSG) to PDF, DICOM and HEIC imaging.
Not part of Textract. Textract reads JPEG, PNG, PDF, and TIFF only and does not convert or render.
Compliance formats
PDF/A conversion and validation for all parts and levels, PDF/UA with automated tagging, encryption and permissions, WCAG 2.2 AA viewer.
Not part of Textract.
Generation, scanning, eSignature
Fluent template-based document generation, Scanbot SDK for mobile document and barcode scanning, Xodo Sign eSignature, and iText (an Apryse company) for Java and .NET PDF programming.
Not part of Textract.
What Our Customers Say
"The speed at which they integrate is superior to their competition and their product roadmap has been good -- they’ve invested in the right things."
Marcus O'Brien
Global Head of Product Management, AutoCAD
"Document integrity is critical. Apryse absolutely solved the problem for us."
Uma Natarajan
CTO, Juume AI
"The rendering performance, combined with how responsive the Apryse engineering team has been, has made a meaningful difference in our product quality."
Amanda Lansman
Sr Product Manager, Dropbox
When to Choose AWS Textract
YOUR PIPELINE ALREADY LIVES IN AWS
If documents land in S3 and the rest of the workflow is Lambda, Step Functions, and IAM, Textract is one permission and one API call away with no new vendor to onboard. Data gravity is real, and fighting it rarely pays.
INVOICES, RECEIPTS, US IDS, OR MORTGAGE PACKETS OUT OF THE BOX
AnalyzeExpense, AnalyzeID, and AnalyzeLending return typed, normalized fields with no setup. Apryse extracts these as generic key-value pairs and a classification label rather than per-type schemas.
QUESTIONS YOU CAN PHRASE IN ENGLISH
Queries let you ask "What is the policy number?" and get an answer with a location and confidence; Custom Queries adapters tune the answers on a handful of your own documents. Apryse has no equivalent. You would pair its structured output with your own LLM.
LOW OR VARIABLE VOLUME OF ENGLISH DOCUMENTS
At $1.50 per 1,000 pages for raw OCR, with a three-month free tier and no license to buy, a prototype or a low-volume workflow over printed English costs almost nothing to start. Pay-per-page is the right economics when volume is small or unpredictable.
FEDRAMP HIGH UNDER AN AWS AUTHORIZATION
If procurement requires a FedRAMP-authorized service boundary rather than software inside your own, Textract carries FedRAMP Moderate in US commercial regions and FedRAMP High in GovCloud, where AWS also commits not to use your content for service improvement.
When to Choose Apryse
DOCUMENTS CAN'T LEAVE YOUR ENVIRONMENT
Regulated, sovereign, or disconnected environments where a shared cloud service is a non-starter, or where "may store and use your inputs unless you opt out" is a policy question you would rather not have. Smart Data Extraction, OCR, and ICR run in-process on your servers, including inside your own AWS account, with nothing sent to Apryse.
YOU'RE SHIPPING A PRODUCT, NOT RUNNING A PIPELINE
When extraction is a feature inside software you sell, you need a component you can embed, version, and license with your product. Textract is a service your product would call; Apryse is a library your product contains.
DOCUMENTS BEYOND SIX WESTERN EUROPEAN LANGUAGES
Textract reads English, French, German, Italian, Portuguese, and Spanish. Apryse's OCR module reads 80+ languages including Chinese, Japanese, Korean, Arabic, Cyrillic, and Devanagari scripts, with the IRIS module for complex CJK layouts.
PREDICTABLE COST AT HIGH VOLUME
Textract meters every feature on top of the last: Forms alone is $50 per 1,000 pages, and Forms with Tables and Queries is $70 before S3, SNS, and Lambda. A package license doesn't grow with page count or feature count. Model your annual volume and feature mix; above a threshold, a license is cheaper and easier to budget.
LONG OR PROTECTED DOCUMENTS WITHOUT ORCHESTRATION
A 300-page contract, drawing set, or filing is processed whole, in one call, including password-protected PDFs. No one-page synchronous limit, no S3 upload, no SNS notification, no reassembling results from an asynchronous job.
YOU NEED THE REST OF THE DOCUMENT STACK
Viewer, annotations, forms, redaction, signatures, editing, conversion, compliance, and a review UI that can overlay extracted fields on the source page, from the same SDK as the extraction. Textract provides none of these.
Backed by a Team of Experts
FAQ
Textract is a cloud API: you send pages to the AWS-operated service and pay per page for each feature you enable. Apryse is an SDK: Smart Data Extraction, OCR, and ICR run inside your own application and infrastructure under a package license, and the same SDK provides viewing, editing, conversion, and compliance capabilities that Textract does not.
Textract has no on-premises, container, or Outposts option; it runs only in 16 AWS regions. Apryse runs on-premises, in private cloud, or fully air-gapped, and also inside your own AWS account, with official guides for Lambda in Node.js and Python and a CloudFormation deployment for WebViewer Server on ECS and Fargate. The difference is that Apryse executes in your compute, not in a shared service.
No. Smart Data Extraction is template-free and works out of the box with generic key-value extraction, table extraction, form-field detection, and classification into 24 predefined categories. There are no Queries, adapters, or custom schemas; teams that need a specific schema pass the structured JSON to an LLM of their choice.
Textract's synchronous operations accept one page and 10 MB; multi-page PDFs must go through S3 and the asynchronous Start and Get operations, which accept up to 500 MB and 3,000 pages and hold results for 7 days. Apryse has no documented per-document cap in licensed use. Processing is local and bound by your hardware.
By default it may. AWS states that Textract "may store and use document and image inputs processed by the service solely to provide and maintain the service and to improve and develop the quality of Amazon Textract and other Amazon machine-learning/artificial-intelligence technologies," and that some content may be stored in another region. You can opt out with an AWS Organizations AI-services opt-out policy; GovCloud regions are exempt, and Custom Queries training data is deleted after training. Apryse never receives your documents.
At low or variable volume, or for raw OCR of English documents, yes, $1.50 per 1,000 pages with a three-month free tier and no license to buy. The picture changes as features stack: Forms is $50 per 1,000 pages, Forms with Tables $65, and Forms, Tables, and Queries $70, before the S3, SNS, and Lambda costs of the asynchronous workflow. Apryse is a package license whose cost does not scale with page count or feature count, with the extraction and OCR modules quoted on request. Model your annual volume and feature mix before deciding.
Bedrock Data Automation is AWS's generative document service, generally available since March 2025: Markdown, HTML, and CSV output in reading order, blueprint-defined custom fields with confidence and page numbers, summaries, and semantic splitting of files up to 3,000 pages, at $0.010 per page for standard output and $0.040 per page for custom output. It runs in eight regions, reads the same six languages as Textract, and is cloud-only. Bedrock Knowledge Bases offers it, not Textract, as a parser. If you are evaluating AWS for generative extraction, compare Apryse with Bedrock Data Automation on the same deployment, language, and determinism axes; this page compares Textract.
Yes. That is the larger part of the SDK. Viewing and annotation of 30+ formats, forms, redaction, digital signatures, PDF and in-browser DOCX editing, conversion between PDF, Office, HTML, CAD, and email, PDF/A and PDF/UA compliance, document generation with Fluent, and mobile scanning with Scanbot SDK all ship under the same Apryse license. Textract is limited to OCR, layout, tables, forms, queries, signatures, and its three specialty parsers.
