macrostack
Browse

The AI stack

Categories

Local & Sovereign AINotes & KnowledgeObservability & MonitoringPassword ManagersWeb AnalyticsTeam ChatSmart HomeNetworking & RoutersVideo ConferencingCloud Storage & SyncPhotos & MediaAPI DevelopmentImage EditingWorkflow Automation & iPaaSDeveloper Tools & ContainersOffice & Productivity SuitesNo-Code DatabasesCode Hosting & Git ForgesProject ManagementEmail Marketing & NewslettersScheduling & BookingError Tracking & Exception MonitoringLog Management & SIEMVPN & PrivacyEmail & Secure MailVector Databases & AI SearchLLM & Agent FrameworksDomains & Web HostingData Removal & PrivacyAuthentication & IdentityHelp Desk & Customer SupportCloud & VPSKubernetes & Container PlatformsEmbedding ModelsPDF & DocumentsAI Coding AssistantsAI Voice & SpeechLLM Observability & EvaluationLLM Gateways & RoutingCloud GPU & AI ComputeCI/CD & build automationData & pipeline orchestrationModel serving & inferenceAI agent frameworksBackend as a serviceSecrets managementFeature flags & experimentationProduct analyticsSearch infrastructureUptime & status monitoringAffiliate & partner platformsVisitor identification & personalisationWikis & internal docsIdentity & access managementData warehouses & analytics enginesCustomer data platformsCRMObject storageBI & dashboardsE-signatureWhiteboards & diagrammingIn-memory data stores & cachingPlatform as a serviceTransactional & bulk emailHeadless CMSDesign & prototypingE-commerce platformsInternal tools & admin panelsManaged databasesForms & surveysFine-Tuning & Model TrainingRAG & Retrieval PlatformsLLM Evaluation & TestingAI Guardrails & Content SafetySpeech Recognition & TranscriptionExperiment Tracking & ML OpsDocument AI & OCR

About

How we rank & score
Migration guide · Document AI & OCR

The 5 best AWS Textract alternatives

Textract is AWS's document-analysis service. Beyond plain OCR it extracts structured data — tables with their cell relationships, form key-value pairs, signatures, and specialised handling for invoices, receipts and identity documents. It is the default answer for turning scanned paperwork into records inside an AWS environment, and its table extraction is genuinely strong on documents that defeat simpler tools.

94
Bottom line

Docling is our pick for most teams — IBM's document converter is MIT licensed, layout-aware, and outputs clean Markdown or JSON that drops straight into a RAG pipeline, which is what most people actually want from this category. Unstructured is the better choice when you need breadth of input formats over depth on any one, since it handles email, HTML, Office documents and more through one interface. PaddleOCR has the strongest raw OCR engine of the group and the best non-Latin script support, Tesseract remains the dependable baseline that runs anywhere, and Marker produces the best Markdown from complex PDFs — with a licence condition worth checking before commercial use.

Jump to the full comparison →

The cost

Per-page usage pricing that varies sharply by feature: plain text detection is around $1.50 per 1,000 pages, while table and form analysis runs closer to $15–$65 per 1,000 pages depending on the combination, with volume tiers. A free tier covers the first months at low volume. The feature multiplier is what surprises people — the analysis you actually need is ten to forty times the price of the OCR you were quoting.

Why people consider an alternative

Per-page pricing collides badly with the most common document-AI job: ingesting a corpus. A one-off run over a document archive for a RAG pipeline is exactly the workload that turns into a four-figure invoice before anyone has evaluated whether the extraction was good enough — and if the chunking was wrong you pay again to redo it. The second reason is iteration: getting document extraction right takes several passes over the same corpus, and per-page billing makes experimentation expensive precisely when you most need to experiment. Third is data residency — documents are frequently the most regulated asset an organisation has, and contracts, medical records and financial statements often cannot be sent to a cloud API regardless of the encryption story.

When AWS Textract is still the right call

Stay for the specialised document types, where Textract's purpose-built models for invoices, receipts and identity documents are meaningfully ahead of anything general-purpose — if you are processing expense claims at volume, that gap is the product. Stay if you are already on AWS and the compliance story matters more than the per-page cost, particularly where a signed BAA or an existing data-processing agreement already covers it. Stay for handwriting, which remains genuinely hard and where the managed services still lead. And stay at low, steady volume where the bill is noise. The tools below are for corpus ingestion, iteration-heavy work, and documents that cannot leave your network.

AlternativeLicenseSelf-hostPricingSovereignty
DoclingMITYesFree, MIT. Runs on your own hardware, CPU or GPU.94
TesseractApache-2.0YesFree, Apache-2.0.95
PaddleOCRApache-2.0YesFree, Apache-2.0.92
UnstructuredApache-2.0YesOpen library free under Apache-2.0; a paid hosted API is offered separately.90
MarkerGPL-3.0 with commercial revenue conditionYesFree below the maintainer's revenue threshold; a commercial licence is required above it.70
94
Macrostack's top pick

Docling

IBM's document converter. Layout-aware PDF to clean Markdown, MIT.

Which one fits your situation?

There is no single best answer — it depends on what you are actually trying to do.

Just tell me the best AWS Textract alternative

Docling94

Our top pick — the highest Sovereignty Score at 94, and the one we would choose ourselves.

Docling vs Tesseract

I have no budget — is there a genuinely free one?

Tesseract95

Open source under Apache-2.0. Free, Apache-2.0.

Tesseract vs Docling

I want to self-host it and own the data

PaddleOCR92

Self-hostable, so your data stays on your own server. Free, Apache-2.0.

PaddleOCR vs Docling

It is just me — I do not want to run a server

Unstructured90

Offers a hosted option, so you can start without infrastructure. Open library free under Apache-2.0; a paid hosted API is offered separately.

Unstructured vs Docling

Every alternative, compared

#1★ TOP PICK

Docling

IBM's document converter. Layout-aware PDF to clean Markdown, MIT.

94
OPEN SOURCEMITSELF-HOSTLOCAL-FIRST

Docling parses PDFs, Office documents, images and HTML into a structured representation that preserves reading order, tables, figures and headings, then exports to Markdown or JSON. It uses purpose-trained layout and table models rather than heuristics, which is why it holds up on multi-column academic papers and financial statements where simpler extractors interleave columns into nonsense. It integrates directly with LlamaIndex and Haystack, is MIT licensed, and runs entirely locally including on CPU.

Strengths

  • +Layout-aware — preserves reading order, tables and structure
  • +Outputs clean Markdown/JSON that drops into a RAG pipeline
  • +Direct integrations with LlamaIndex and Haystack
  • +MIT, fully local, no per-page cost

Trade-offs

  • Slower per page than cloud OCR on very large batches
  • Handwriting support is weak compared with Textract
  • No specialised invoice or receipt models
Free, MIT. Runs on your own hardware, CPU or GPU.
#2

Tesseract

Thirty years old, Apache-2.0, and it runs absolutely everywhere.

95
OPEN SOURCEApache-2.0SELF-HOSTLOCAL-FIRST

Tesseract is the OCR engine most of the world's text extraction has quietly gone through, originally from HP, developed by Google for a decade, and Apache-2.0 throughout. It recognises over a hundred languages, has bindings for every language you might be writing in, and is packaged in every Linux distribution. Modern neural extractors beat it on complex layouts, but on clean scanned text it is fast, dependable and effectively free of dependencies — often the right tool precisely because it is boring.

Strengths

  • +Over a hundred languages, bindings for everything
  • +Extremely mature and stable — decades of production use
  • +Packaged everywhere; trivial to install and deploy
  • +Apache-2.0 with no dependency weight

Trade-offs

  • Weak on complex layouts, multi-column pages and tables
  • Needs image pre-processing for good results on poor scans
  • No document understanding — it gives you text, not structure
Free, Apache-2.0.
#3

PaddleOCR

The strongest open OCR engine, and the best at non-Latin scripts.

92
OPEN SOURCEApache-2.0SELF-HOSTLOCAL-FIRST

PaddleOCR is a comprehensive OCR toolkit covering text detection, recognition, table extraction, layout analysis and key-value extraction, with support for around eighty languages. Its non-Latin script handling — Chinese, Japanese, Korean, Arabic — is clearly the best of the open options, and it ships lightweight mobile-scale models alongside the accurate server ones. Apache-2.0. If raw recognition accuracy on difficult scans is the constraint, this is the engine.

Strengths

  • +Best-in-class open recognition accuracy on difficult scans
  • +Around eighty languages, with excellent non-Latin coverage
  • +Includes table, layout and key-value extraction
  • +Lightweight models for edge and mobile deployment

Trade-offs

  • Built on the PaddlePaddle framework — an extra dependency to adopt
  • Documentation is stronger in Chinese than in English
  • Output needs more post-processing than Docling's Markdown
Free, Apache-2.0.
#4

Unstructured

One interface for every document format you will actually be handed.

90
OPEN SOURCEApache-2.0SELF-HOSTLOCAL-FIRST

Unstructured normalises an unusually wide range of inputs — PDF, Word, PowerPoint, Excel, HTML, email, EPUB, images — into a consistent element structure of titles, narrative text, tables and lists, ready for chunking. Breadth is the point: real corpora are never one format, and writing a parser per type is where ingestion projects stall. The open library is Apache-2.0 and runs locally; the company also sells a hosted API with additional models.

Strengths

  • +Widest input-format coverage of anything here
  • +Consistent element output regardless of source format
  • +Chunking strategies built in for RAG pipelines
  • +Apache-2.0 open library

Trade-offs

  • Best-quality models sit in the paid hosted tier
  • Local install pulls in heavy system dependencies
  • Depth on complex PDFs is below Docling's
Open library free under Apache-2.0; a paid hosted API is offered separately.
#5

Marker

The best PDF-to-Markdown quality here — check the licence first.

70
SOURCE-AVAILABLEGPL-3.0 with commercial revenue conditionSELF-HOSTLOCAL-FIRST

Marker converts PDFs to Markdown with the highest fidelity of anything in this list: it handles equations, code blocks, tables and multi-column layouts, and optionally uses an LLM pass to improve difficult sections. For academic papers and technical documents the output quality is noticeably ahead. The caveat we would rather state than have you find in a licence review: Marker is GPL-3.0 with an additional revenue condition from its maintainer, so organisations above a revenue threshold need a commercial licence. Free for research and for smaller organisations, but not unconditionally.

Strengths

  • +Best PDF-to-Markdown fidelity of the options here
  • +Handles equations, code blocks and complex tables well
  • +Optional LLM pass for difficult pages
  • +Runs entirely locally

Trade-offs

  • Not unconditionally open source — revenue-gated commercial terms
  • GPL-3.0 copyleft affects how you can distribute derived work
  • GPU strongly recommended for reasonable throughput
Free below the maintainer's revenue threshold; a commercial licence is required above it.

Questions people ask

What is the best open-source alternative to AWS Textract?

Docling for most teams — MIT licensed, layout-aware, and it outputs clean Markdown or JSON that drops straight into a RAG pipeline, which is what this category is usually bought for. If your corpus spans many formats rather than PDFs alone, Unstructured handles the widest input range through one interface.

Why does per-page pricing hurt so much on document AI?

Because the defining workload is a corpus ingest, and getting extraction right takes several passes over the same documents. At $15 to $65 per thousand pages for table and form analysis, a hundred-thousand-page archive is a four-figure bill — and you pay it again each time you change the chunking strategy. Local extraction makes iteration free, which is the difference between tuning your pipeline and guessing at it.

Is open-source OCR accurate enough?

For printed text, yes — PaddleOCR and Docling are competitive with the cloud services on clean scans, and Docling's layout handling on complex documents is genuinely strong. The remaining gaps are handwriting, where the managed services still lead clearly, and specialised document types like invoices and identity documents, where Textract's purpose-trained models are ahead of anything general-purpose.

What is the difference between OCR and document AI?

OCR gives you the characters. Document AI gives you the meaning: which text is a heading, which cells belong to which table row, which value pairs with which form label, what the reading order is on a two-column page. For RAG that distinction decides everything — a table flattened into a line of numbers is worse than useless to a retriever, because it retrieves confidently and answers wrongly.

Is AWS Textract bad?

No. Its table extraction is excellent, the specialised invoice, receipt and identity-document models have no real open equivalent, and handwriting is meaningfully better than anything you can self-host. If you are already on AWS with the compliance paperwork in place and volume is steady, it is a defensible buy. The concerns are per-page cost on corpus ingestion, expensive iteration, and documents leaving your network.

Can I run document extraction with no GPU?

Yes. Tesseract and Docling both run on CPU, and for a few thousand pages that is entirely reasonable — you are trading wall-clock time for a bill you do not pay. A GPU helps materially on large corpora and is close to required for Marker at volume, but for most first ingests a laptop overnight is the honest comparison against a cloud invoice.

Compare them head-to-head

Related comparisons

Entry last verified 2026-08-11. Licenses and pricing change — spotted something out of date? That's a correction we want.

The Macrostack brief

New swaps, worth your inbox.

A short, occasional email when we add a high-intent alternative or ship a new head-to-head. No spam, no selling your address — unsubscribe in one click.