Amazon Textract Alternative for PDF to Excel: AWS Textract Table Extraction Without Code or Per Page Pricing
Amazon Textract is an AWS API. It returns JSON, not a spreadsheet, and getting a multi page PDF through it means an S3 bucket, an asynchronous job, and a script that walks the response blocks. If the spreadsheet was the whole point, drop a file on the right and look at the output first.
OCR on scanned pages is included on paid plans, and files are deleted after processing.
The converter is not live yet
We are still building the conversion engine, so we are not accepting files or payments right now. Leave us your address and we will write to you the day it goes live.
No charge, no account needed. We only email you once, at launch.
Not accepting files yet. We are finishing the conversion engine.
The short answer
Amazon Textract is a developer API for document extraction. It charges per page per feature, with table extraction listed at $15 per 1,000 pages and forms at $50 per 1,000 pages in US West (Oregon), and those charges stack when you request both. It outputs JSON blocks, so producing an Excel file requires code you write yourself. PDFXLSX is the alternative for people who wanted the spreadsheet rather than the pipeline: upload a PDF in a browser, download XLSX or CSV, on a flat plan from $9 a month with no AWS account involved.
Last updated August 2026. Amazon Textract rates, quotas and free tier taken from the AWS Textract pricing page and the Textract developer guide, checked this month. AWS prices vary by region and change without notice, so confirm on the AWS pricing page before you budget.
XLSX
Not JSON blocks
$9/mo
Flat, published
0
Lines of code
OCR
Included on paid plans
What Amazon Textract actually is
Textract is a machine learning service inside AWS that reads documents. You call it from the SDK or the CLI, it detects text, tables, form key value pairs and signatures, and it hands back a JSON response made of Block objects: pages, lines, words, tables, cells, and the relationships between them. It is a good service and a lot of production document pipelines run on it.
The friction shows up when the person who needs the data is not an engineer. There is no button in the AWS console that turns a folder of statements into spreadsheets. There is a console demo for a single file, and AWS publishes Python samples plus the open source Textractor library that can write tables to CSV or Excel, but that is code somebody has to own, deploy and maintain.
There is also a quota that surprises people. Synchronous calls accept a PDF of exactly one page. Anything longer has to go through the asynchronous API, which means uploading the file to S3, starting a job, polling for completion, then paging through the results. That is the real cost of Textract, and it has nothing to do with the per page rate.
Amazon Textract hard limits
Published set quotas from the Textract developer guide. These cannot be raised.
- 1Synchronous operations: 10 MB, and PDF or TIFF limited to a single page.
- 2Asynchronous operations: PDF and TIFF up to 500 MB and 3,000 pages, and the file must be in S3.
- 3Accepted formats are JPEG, PNG, PDF and TIFF. XFA based PDFs are not supported.
- 4PDFs cannot be password protected, and pages cannot exceed 40 inches or 9,000 points.
- 5Text detection covers English, French, German, Italian, Portuguese and Spanish. Handwriting is English only.
- 6Vertical text is not supported, and text under 15 pixels tall (about 8 point at 150 DPI) is below the detection floor.
Amazon Textract vs PDFXLSX
An honest comparison, including the rows where Textract is plainly the stronger choice. One is cloud infrastructure you build on, the other is a finished tool.
| Capability | Amazon Textract | PDFXLSX |
|---|---|---|
| What it is | An AWS API for document text, table, form and signature detection | A browser tool that turns PDF tables into Excel or CSV |
| What you get back | JSON Block objects. Excel or CSV requires the AWS sample scripts or the Textractor library | An XLSX or CSV file you download and open |
| Setup before first result | AWS account, IAM permissions, an S3 bucket for anything over one page, SDK code and job polling | Upload a file on this page |
| Multi page PDF | Asynchronous only. Up to 3,000 pages and 500 MB, but the file must sit in S3 first | Upload it directly, including long documents |
| How you are billed | Per page, per feature, and the features stack. Tables listed at $15 per 1,000 pages, Forms at $50 | Flat monthly plan from $9, billed per conversion rather than per page |
| Throughput ceiling | Effectively none. Built for millions of pages with a documented volume discount above 1M | Unlimited conversions on the Pro plan, but this is a tool, not a processing platform |
| Compliance and data residency | HIPAA eligible, and in scope for PCI, ISO and SOC. Data stays in your AWS account and chosen region | Encrypted transfer and deletion after processing, but no SOC 2 report, no SSO and no HIPAA BAA |
| Beyond tables | Form key value pairs, natural language Queries, signatures, and specialised Expense, ID and Lending models | Tables and text into a spreadsheet. No decisioning, no field level rules |
| Handwriting | Supported, English only | OCR is tuned for printed tabular documents rather than handwriting |
| Best fit | Engineering teams embedding extraction in an application or a high volume pipeline | Accountants, analysts and operations teams who need clean spreadsheets without building anything |
Read that honestly and Textract wins four rows outright: throughput ceiling, compliance and data residency, everything beyond table extraction, and handwriting. Those matter. If you are processing hundreds of thousands of pages a month, or you need a HIPAA business associate agreement, or the data cannot leave your own AWS account, buy Textract and stop reading here. This page argues only that a large share of the people pricing Textract wanted an Excel file, not a data pipeline.
Amazon Textract pricing, and the part that catches people out
Textract has no subscription and no minimum. You pay for the pages you send, and the rate depends on which feature you request. These are the published US West (Oregon) rates as of August 2026. Rates differ by region, so check yours.
| Textract feature | First 1M pages / month | Above 1M pages | What it gives you |
|---|---|---|---|
| Detect Document Text | $1.50 per 1,000 | $0.60 per 1,000 | Plain OCR. Words and lines, no table structure |
| Tables | $15.00 per 1,000 | $10.00 per 1,000 | Rows, columns, cells and merged cell relationships |
| Forms | $50.00 per 1,000 | $40.00 per 1,000 | Key value pairs on structured forms |
| Queries | $15.00 per 1,000 | $10.00 per 1,000 | Ask a plain question of a page, up to 30 per page async |
| Layout | Included with Tables | Included with Tables | Reading order, headers, paragraphs, titles |
| Analyze Expense | $10.00 per 1,000 | $8.00 per 1,000 | Invoice and receipt fields plus line items |
| Analyze Lending | $70.00 per 1,000 | $55.00 per 1,000 | Mortgage document classification and splitting |
The charges stack. If you call AnalyzeDocument with both TABLES and FORMS on the same page, you are billed for both, so that page costs $0.065 rather than $0.050. It is an easy mistake to make when you copy a code sample that enables every feature, and on 20,000 pages a month it is the difference between a $300 bill and a $1,300 bill.
The free tier does not cover tables. Textract's free tier runs for three months and includes 1,000 pages a month of Detect Document Text, 1,000 pages of Analyze Document limited to signatures, 100 pages each of Analyze Expense and Analyze ID, and 2,000 pages of Analyze Lending. Table extraction is not in it, so the moment you test the thing you actually came for, you are paying.
Textract bills per page, we bill per file. That distinction decides most real comparisons. A 42 page bank statement is 42 billed pages on Textract Tables, about $0.63, plus whatever the surrounding code costs to run. Here it is one conversion. Fifty statements of that length is roughly 2,100 pages, around $31 a month of Textract Tables, against $9 on our Starter plan. Flip the volume to 200,000 pages and Textract at $3,000 is the wrong shape of question: at that scale you want the pipeline, and the pipeline is worth building.
A note on the wider cloud OCR market
Textract's closest equivalent is Azure AI Document Intelligence, and buyers usually price both. Microsoft's free F0 tier allows 500 pages a month on an ongoing basis rather than for 90 days, and third party pricing summaries put the Layout model and the prebuilt models around $10 per 1,000 pages on the standard tier. We are quoting that from secondary sources because Microsoft renders its own rate table dynamically and it does not appear in the page source, so treat it as indicative and confirm in the Azure pricing calculator. Both services share the same shape: cheap per page, and an engineering project before a single spreadsheet exists. Our full breakdown of what a PDF to Excel converter costs compares the four pricing models the market uses, and the Amazon Textract pricing guide works the per page math out in detail.
The documents people build Textract pipelines for
In most of these cases the goal was a spreadsheet, and the pipeline was a means to it.
Bank statements
Long, repetitive, and the worst possible fit for per page billing. Every transaction row has to survive intact, including negatives and running balances.
Invoices
Textract has a dedicated Analyze Expense model for these. If you only need the line items in a sheet for a supplier review, the general table path is cheaper.
Trial balances and ledgers
Accounting exports that arrive as PDF because the previous system only printed. Column alignment matters more than field detection.
Rent rolls and operating statements
Property reports where the layout changes with every management company, which is exactly where a fixed template breaks and general table detection wins.
Scanned and faxed documents
No text layer at all, so OCR has to run first. Both tools handle it; only one of them needs you to decide which API to call.
Purchase orders
Structured enough to look like a Forms job, tabular enough that Tables usually answers it for a third of the price.
How the conversion works here
Upload the PDF
Drag one file or a batch into the box at the top of this page. No bucket, no upload step, no page count to check first. There is a batch converter if you have a folder of them.
The table gets rebuilt
A PDF stores characters at coordinates and has no concept of a cell, so columns are inferred from position. Scanned pages get OCR first. Numbers arrive as numbers, and negatives stay negative rather than landing as text in parentheses.
Download XLSX or CSV
Open it in Excel, Google Sheets or Numbers. If something looks off, our guide on checking a conversion for errors covers the five checks worth running before you trust the numbers.
When Amazon Textract is the better choice
There is a real line here and it is worth drawing clearly, because sending the wrong buyer to the wrong tool wastes everybody's month.
Extraction lives inside your product
If your own application ingests customer documents and the output feeds a database rather than a person, you need an API. Textract is a strong one, it is in the same account as the rest of your infrastructure, and the per page rate at scale is hard to beat.
Regulated data that cannot leave your account
Textract is HIPAA eligible and in scope for PCI, ISO and SOC, and the documents stay in your own AWS region under your own retention policy. We do not offer a business associate agreement or a SOC 2 report, so for protected health information the answer is Textract.
You need more than tables
Form key value pairs, natural language Queries against a page, signature detection, driver licence parsing, mortgage packet classification: these are separate Textract models with no equivalent here. If the job is reading a document rather than reproducing a table, that is the product.
Volume in the hundreds of thousands of pages
Above a million pages a month the rate drops again and the economics get genuinely good. At that scale nobody is clicking upload anyway, and an enterprise conversion workflow is a build, not a purchase.
Amazon Textract alternative: common questions
Can Amazon Textract export tables to Excel?
Not on its own. Textract returns JSON Block objects, and turning those into a spreadsheet is code you run yourself. AWS publishes Python samples for exporting tables to CSV, and the open source Textractor library has a to_excel method that preserves merged cells. The console will also let you download CSV for a single test document, which is fine for a one off and not a workflow.
How much does Amazon Textract cost per page?
In US West (Oregon), plain OCR is $0.0015 a page, table extraction is $0.015 a page, and form key value extraction is $0.050 a page for the first million pages each month. Features are billed additively, so requesting tables and forms together costs $0.065 a page. Above one million pages the rates fall to $0.0006, $0.010 and $0.040 respectively.
Is Amazon Textract free?
There is a free tier, but it lasts three months and it excludes the feature most people want. It covers 1,000 pages a month of Detect Document Text, 1,000 pages of Analyze Document restricted to signatures, 100 pages each of Analyze Expense and Analyze ID, and 2,000 pages of Analyze Lending. Table extraction is billed from the first page.
Can I use Amazon Textract without an AWS account?
No. Textract is an AWS service, so it needs an account, IAM credentials and, for any document longer than one page, an S3 bucket to hold the file while the asynchronous job runs. If setting that up is the blocker rather than the price, a browser converter removes the whole layer: you upload a file and download a spreadsheet.
What are the page limits for Amazon Textract?
Synchronous calls accept 10 MB and a single page of PDF or TIFF. Asynchronous calls accept PDF and TIFF up to 500 MB and 3,000 pages, with the file stored in S3. That one page synchronous limit is the quota that surprises most people, because it means even a three page invoice needs the asynchronous path and everything that comes with it.
How does Amazon Textract compare to Azure Document Intelligence?
They solve the same problem from inside two different clouds, and the choice usually follows whichever one you already run. Azure's free F0 tier allows 500 pages a month with no 90 day expiry, which makes evaluation easier, and third party summaries put its Layout and prebuilt models near $10 per 1,000 pages. Both still hand you structured data rather than a finished spreadsheet.
Is Amazon Textract accurate on tables?
It is strong on clean, ruled tables and it reports merged cell relationships, which many extractors do not. Borderless financial tables, multi line cells and continuation rows across page breaks are harder for every tool including this one. Whatever you evaluate, run your worst document rather than your cleanest, because that is the one that decides how much manual cleanup you inherit.
What are the best Amazon Textract alternatives?
It depends which half of the job you need. For an API inside a pipeline, the comparable services are Azure AI Document Intelligence, Google Document AI and document AI vendors such as Nanonets and Docsumo. For getting tables into spreadsheets without writing code, the comparison set is converters: see the best PDF to Excel converter roundup, or the PDFTables alternative and Docparser alternative pages.
Does Amazon Textract work on scanned documents?
Yes, OCR is the core of the service and it reads handwriting in English as well as printed text. The practical limits are worth knowing: text has to be at least 15 pixels tall, roughly 8 point at 150 DPI, and vertical text is not supported at all. We run OCR on scanned pages automatically on paid plans, with no separate call and no separate charge.
Do I need Textract just to get a table out of a PDF?
Only if the extraction has to happen without a person present. If someone is already opening each document, the bottleneck is retyping rather than routing, and a PDF table extractor removes that work in seconds with nothing to deploy. Revisit an API once the volume genuinely justifies owning the code, and if you would rather build it yourself in the meantime, our walkthrough of extracting tables from PDF with Python shows the open source route.
More PDF to Excel options
Best PDF to Excel Converter: Software and Online Tools
Best PDF to Excel converter compared: 9 real tools ranked on accuracy, OCR for scans, batc...
Smallpdf Alternative for PDF to Excel and CSV
Smallpdf alternative focused on PDF to Excel and CSV. Auto-detects tables, runs OCR on sca...
iLovePDF Alternative for PDF to Excel and CSV
iLovePDF alternative focused on PDF to Excel and CSV. Auto-detects tables, runs OCR on sca...
PDF-XChange Editor PDF to Excel Alternative, No Install
PDF-XChange Editor keeps its Excel export locked until you pay about $62. Convert PDF tabl...
See the spreadsheet before you write the pipeline
Take one of the documents you were about to send through Textract and convert it here first. If the output is all you needed, you just saved yourself a sprint.
Try the converter