The short answer: the best OCR API for tables depends on what you need back. If you want table structure in JSON at the lowest price, Azure Document Intelligence Layout and Google Document AI Layout Parser both cost $10 per 1,000 pages, and Amazon Textract Tables costs $15. If you want a finished Excel or CSV file instead of JSON, the choice narrows to Adobe PDF Extract, which is quote only above its free tier, or a flat priced converter API such as ours, which is included on the Pro plan at $199 a month with no page meter.
Every price below comes from the vendor's own pricing page or public price list, in US dollars, pay as you go. We sell one of the options on this list, so we have kept our own entry to the facts and said plainly where the cloud APIs are cheaper. They usually are, per page. Whether they are cheaper per finished spreadsheet is a different question, and it is the one this comparison is built to answer.
Best OCR API for tables by price per page
This table only covers the endpoint of each service that returns tables. A plain OCR endpoint that returns lines of text is cheaper on every platform, but it leaves you to rebuild columns from word coordinates yourself.
| API | Table endpoint | List price | Returns | Best for |
|---|---|---|---|---|
| Azure Document Intelligence | Layout model | $10 per 1,000 pages | JSON cells with row and column indexes, or Markdown | Teams already on Azure |
| Google Document AI | Layout Parser / Form Parser | $10 / $30 per 1,000 pages | JSON Document object | Teams already on Google Cloud |
| Amazon Textract | AnalyzeDocument, Tables | $15 per 1,000 pages (first 1M a month) | JSON blocks with cell relationships | Teams already on AWS |
| Mindee | Extraction models | $29 a month platform fee plus credits at about $0.05 to $0.06 a page | JSON fields and line items | Developers who want a hosted model without a cloud account |
| Nanonets | Data Extraction AI block | 0.30 credits a page, about $0.30 at the Starter credit price | JSON, or exports to connected apps | Document workflows with approvals and exports |
| Veryfi | Document type APIs | $0.16 an invoice, $0.25 a bank statement, $500 a month minimum | JSON fields and line items | High volume receipts, invoices and checks |
| Adobe PDF Services | PDF Extract API | 500 free transactions a month, paid tiers quote only | JSON plus tables as CSV or XLSX files | Enterprises already buying Adobe |
| Google Cloud Vision | Document Text Detection | $1.50 per 1,000 pages after 1,000 free | Text only, no table structure | Plain OCR, not tables |
| pdfxlsx API | /api/v1/conversions | Included on the Pro plan, $199 a month, unlimited conversions | XLSX, CSV, or parsed tables as JSON | Teams that want the spreadsheet, not the parsing work |
Two rows need a warning. Google Cloud Vision is often recommended as an OCR API, and it is a good one, but it does not return tables. People who pick it for invoices or statements end up paying Document AI rates later anyway. And the Mindee, Nanonets and Veryfi prices are not directly comparable with the cloud meters, because they bill by credit, by workflow step or by document type. The cost table further down puts all of them on the same workload.
Which OCR API is best for table extraction?
For raw table structure at volume, Azure Document Intelligence Layout and Google Document AI Layout Parser are the best value at $10 per 1,000 pages, and Amazon Textract is the natural pick inside AWS at $15. All three return JSON, so they suit developers building a pipeline. If the output has to be an Excel file, an API that returns XLSX saves the code that rebuilds rows from cells.
Accuracy on your own documents should break the tie, not a few dollars per thousand pages. Run the same 50 pages through two or three candidates and count the cells you have to correct by hand. On bank statements with running balances, invoices with wrapped descriptions and reports with merged headers, that count varies far more between vendors than the price does.
Azure Document Intelligence
The Layout model returns each table as a list of cells with row and column indexes, spans and a header flag. It is $10 per 1,000 pages on the S0 tier, with prebuilt invoice, receipt, bank statement and US tax form models at the same rate. Microsoft allows up to 2,000 pages and 500 MB per document. The free F0 tier analyzes only the first two pages of each file, so test on S0. Full meter by meter figures are in our Azure Document Intelligence pricing guide, and the case for a no code route is on the Azure Document Intelligence alternative page.
Google Document AI
Layout Parser at $10 per 1,000 pages returns tables with document structure; Form Parser at $30 adds key and value pairs and checkboxes. The practical catch is size: online requests take 15 pages, and batch jobs take up to 100 pages per document on Form Parser. Longer statements have to be split first. See Google Document AI pricing for the processor list and page counting rules.
Amazon Textract
Tables cost $15 per 1,000 pages for the first million pages a month and $10 after that, and Layout comes free with Tables. Features stack: ask for Tables and Forms on the same page and you pay $0.065, not $0.015. The three month free tier does not include tables. Our Amazon Textract pricing breakdown walks through the stacking in detail.
Mindee
Mindee charges a platform fee, $29 a month on Starter or $119 on Pro, plus prepaid page credits: 500 for $30, 5,000 for $265. One page is one credit. Credits expire at the end of the period, and overage is billed at the plan's unit rate plus 20 percent. It is simpler to start with than a hyperscaler, and the Mindee alternative page shows where its costs cross ours.
Nanonets and Veryfi
Both are document automation platforms with an API rather than raw OCR meters. Nanonets bills 0.30 credits per extracted page, then charges again for formatting, lookups and exports, some of them per table row, as our Nanonets pricing guide explains. Veryfi prices by document type, from $0.08 a receipt to $0.25 a bank statement, but paid use starts at a $500 a month commitment.
Adobe PDF Extract API
Adobe is the one cloud API on this list that can hand back tables as CSV or XLSX files alongside its JSON. The free tier is 500 Document Transactions a month, and Extract counts one transaction per five pages. Beyond that there is no published price; you talk to sales and sign a separate agreement. The numbers customers have reported are in Adobe PDF Services API pricing.
pdfxlsx API
Ours is a conversion API, not a general OCR service. You POST a PDF, poll one status endpoint and download XLSX or CSV, or read the parsed tables as JSON. Scanned pages go through OCR automatically, and each table lands on its own sheet with amounts kept numeric. It is included on the Pro plan at $199 a month with unlimited conversions and unlimited team seats. It does not do named field extraction, classification or custom model training, and if you need those, one of the platforms above is the better buy. The endpoints are documented on the PDF to Excel API page.
What 5,000 table pages a month cost on each API
Here is one realistic workload priced across the board: 500 statements and reports of ten pages each, 5,000 pages a month, where the goal is every table in a spreadsheet. List prices, no negotiated discounts.
| API | Monthly bill at list price | What you still have to build |
|---|---|---|
| Azure Document Intelligence, Layout | $50 | JSON to XLSX conversion, retries, storage |
| Google Document AI, Layout Parser | $50 | JSON to XLSX conversion, batch jobs for files over 15 pages |
| Amazon Textract, Tables | $75 | JSON to XLSX conversion, async jobs for multipage PDFs |
| Google Document AI, Form Parser | $150 | As Layout Parser |
| pdfxlsx API, Pro plan | $199 | Nothing; the file is the output |
| Mindee, Starter plus a 5,000 credit bundle | $294 | JSON to XLSX conversion |
| Nanonets, extraction only | About $1,500 in credits | Formatting and export blocks are billed on top |
On this workload the three hyperscalers are cheaper than we are on the invoice, and we would not argue otherwise. The gap is $125 to $150 a month. That is the budget you have for writing and maintaining the step that turns cells into clean rows, handling amounts in parentheses and headers that repeat on every page, and fixing it when a bank changes its layout. At a loaded developer rate, that budget runs out in a couple of hours. The volumes where the meters clearly win are much larger ones, tens of thousands of pages a month, with an engineer who owns the pipeline anyway.
Is Google Cloud Vision good for tables?
No. Cloud Vision's text detection returns words and lines with coordinates, not rows and columns, so any table has to be reconstructed in your own code. It is $1.50 per 1,000 pages after 1,000 free each month, which is why it looks attractive. For tables, Google points you to Document AI instead, where Layout Parser costs $10 per 1,000 pages.
The same trap exists on every platform. Plain OCR is roughly seven to ten times cheaper than the table endpoint, and teams sometimes pick it to save money and then spend weeks writing column detection. If your documents have ruled tables and consistent layouts, that can work. If they come from dozens of banks or vendors, it rarely does.
Which OCR API returns Excel instead of JSON?
Adobe PDF Extract can write tables out as CSV or XLSX, but paid use is quote only. General file conversion APIs such as ConvertAPI convert PDF to XLSX per conversion, from $35 a month for 1,000 conversions, though they are not built around OCR on scans; our ConvertAPI pricing page covers how a conversion is counted. A dedicated PDF to Excel API like ours returns XLSX with OCR included at a flat monthly price.
Tables do not only live in PDFs. When the figures you need sit on supplier portals or public web pages instead, a web scraping API that returns clean structured data is the better tool for that half of the job, and the PDF side stays with an OCR or conversion API.
How to choose an OCR API for tables
Start with where the output goes, not with the price per page.
- The output feeds your own software. Use the table endpoint of the cloud you already run on: Azure Layout, Google Layout Parser or Textract Tables. Build once, pay pennies per page.
- You need named fields, not grids. Invoice totals, vendor names and statement balances are what prebuilt models and platforms like Mindee, Nanonets and Veryfi are for.
- The output is a spreadsheet a person opens. Pay for an API that returns the file, or skip the API entirely and let the team convert PDF to Excel in the browser, up to 20 files per batch.
- You are not sure yet. Send the same 50 pages through two candidates and count the corrections before you commit any code.
If you are in the third group, upload one of your own statements or reports with the converter at the top of this page and compare the sheet with the original before you choose a plan. Plans and limits are on the pricing page.