Is AWS Textract free? What it actually costs in 2026
Textract's free tier lasts three months: 1,000 pages a month of plain text detection, 100 pages of forms, tables and queries, 100 expense and 100 ID pages. After that, text detection is $0.0015 a page and forms $0.05.
The free tier is a launch allowance for new AWS customers, not a permanent floor. Detect Document Text: 1,000 pages a month; Analyze Document features: 100 pages a month each; Analyze Expense and ID: 100 pages; Analyze Lending: 2,000. Custom Queries have no free tier.
What the free tier actually includes
- Three months only, new AWS accounts
- 1,000 pages a month of text detection; 100 pages a month for forms, tables, queries
- No free tier for Custom Queries
Where the wall is
Month four, or a real batch. At $0.0015 a page plain OCR stays cheap; forms at $0.05 and tables at $0.015 a page are what make a 100,000-document backlog a $5,000 line, which is where open OCR on your own hardware is compared.
After that: US West (Oregon), first million pages: Detect Document Text $0.0015 a page ($0.0006 after 1M); tables $0.015; forms $0.05; queries $0.015; tables + forms + queries $0.070; Analyze Expense $0.01; Analyze ID $0.025 a page.
Can you self-host AWS Textract? No — cloud only
Textract is an AWS API. Docling, Unstructured and PaddleOCR are the open-source document-extraction tools compared here and run locally.
What it actually takes to run one yourself
Real RAM against the documented minimum, monthly cost, setup time, and the one thing that catches people out.
Self-hosting Docling
Self-hosting Unstructured
When paying 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.
What you are locked into
Low. Extraction is stateless: document in, structured output out. Nothing accumulates. What needs rewriting is the parsing code that consumes Textract's specific block-and-relationship JSON, which is more idiosyncratic than most — the alternatives here emit Markdown or simpler structures, so the consuming code usually gets shorter. Budget days.
If you would rather not pay: Docling
Docling
IBM's document converter. Layout-aware PDF to clean Markdown, MIT.
Free, MIT. Runs on your own hardware, CPU or GPU.
See all 5 AWS Textract alternatives comparedCompare the free options head-to-head
Read from Amazon Web Services's own pricing and documentation on 2026-09-26 — 25 days ago. Vendors change prices without notice; confirm before you commit.