AWS charges $0.010 a page for standard output and $0.040 a page for a custom schema, so $10.00 and $40.00 per 1,000 pages. Contrary to what most people assume about a newer service, that makes Bedrock Data Automation the expensive way to do OCR on AWS, not the cheap one.
Upload a document to extract
Drop files here or click to upload
Up to 50 files
Uploading...
A per-page rate cannot tell you whether the fields come out right on your documents. Drop one in and look at the output before you commit to a vendor.
Amazon Bedrock Data Automation charges $0.010 per page for standard output, which is $10.00 per 1,000 pages, and $0.040 per page for custom output against a blueprint you define, which is $40.00 per 1,000 pages. Blueprints wider than 30 fields add $0.0005 per field per page. Audio is $0.006 a minute and video is $0.050 a minute. There is no volume discount at any scale. The comparison that actually matters is with Amazon Textract, which charges $1.50 per 1,000 pages to pull plain text off the same page, so Bedrock Data Automation is about 6.7 times more expensive for straight OCR. It earns its rate only on structured extraction, where its $40.00 undercuts Textract Forms at $50.00 and the Forms plus Tables plus Queries bundle at $70.00.
You are paying for the schema, not for better OCR. Price the job accordingly.
AWS publishes these per page and per minute. The right-hand column normalizes them onto the unit the rest of the industry quotes, per 1,000 pages, so you can put them beside Textract, Azure and Google without doing the arithmetic yourself.
| Meter | What it covers | Unit | Official rate | Per 1,000 pages |
|---|---|---|---|---|
| Documents, standard output | Text, summary, tables and figure captions in a fixed shape | Per page | $0.010 | $10.00 |
| Documents, custom output | Your own schema, blueprint of up to 30 fields | Per page | $0.040 | $40.00 |
| Each blueprint field above 30 | The surcharge for a wider schema | Per field, per page | $0.0005 | +$0.50 per field |
| Images, custom output | Your own schema, blueprint of up to 30 fields | Per image | $0.005 | Not page based |
| Audio, standard output | Transcription and speaker analysis | Per minute | $0.006 | $0.36 per hour |
| Video, standard output | Scene detection, transcript and description | Per minute | $0.050 | $3.00 per hour |
Read from the AWS Bedrock pricing page on July 13, 2026. One note on what AWS does not publish: the table gives a custom output rate for images ($0.005 per image) but no clearly stated standard output rate for images. We would rather say that than invent a number, so check the console before you budget image work.
Bedrock Data Automation is the only one of the three hyperscaler extraction services that charges you more for asking for more fields. Past thirty fields in a blueprint, AWS adds $0.0005 per field per page. AWS gives the example itself: a 40-field blueprint costs $0.045 per page rather than $0.040.
That sounds trivial until you push it. The blueprint maximum is 100 leaf fields. Take it there and you are paying $0.040 plus 70 extra fields at $0.0005, which is $0.075 per page, or $75.00 per 1,000 pages. The headline rate has nearly doubled, and the job now costs more than Textract's most expensive bundle. Azure and Google do not care how many fields your schema asks for; the rate is the rate.
The practical advice is simple. Count the fields in your blueprint before you model the cost, and if you are near the line, ask whether you really need field thirty-one.
The base rate, no surcharge. That is $0.040 per page.
AWS's own worked example. That is $0.045 per page.
30 extra fields at $0.0005. That is $0.055 per page.
70 extra fields at $0.0005. That is $0.075 per page.
Right-hand figures are per 1,000 pages. A blueprint can hold at most 100 leaf fields, of which no more than 30 may be list fields.
For plain text, no, and it is not close. For a form with a schema, yes. Both sets of rates come off the AWS pricing pages, checked the same day. The honest summary is that these are different tools with a small and specific overlap, and the overlap is where BDA wins.
| The job | Bedrock Data Automation | Amazon Textract | Winner |
|---|---|---|---|
| Plain text off a scanned page | $10.00 (standard output) | $1.50 (Detect Document Text) | Textract, 6.7x cheaper |
| Key and value pairs from a form | $40.00 (custom output) | $50.00 (Analyze Document Forms) | Bedrock Data Automation, 20% cheaper |
| A handful of specific answers | $40.00 (custom output) | $15.00 (Queries) | Textract, 2.7x cheaper |
| Forms and tables and queries together | $40.00 (custom output) | $70.00 (bundled) | Bedrock Data Automation, 43% cheaper |
| An invoice into structured fields | $40.00 (custom output) | $10.00 (Analyze Expense) | Textract, 4x cheaper |
| A 100-field schema | $75.00 (the 30-field surcharge) | $70.00 (bundled, no field cap) | Textract, and the gap inverts |
| Volume discount above 1M pages | None at any volume | $0.60 per 1,000 (Detect) | Textract |
| Max pages in one document | 3,000 with the splitter | 3,000 (async PDF) | Tie |
| Handwriting | Yes | Yes | Tie |
| CJK and vertical text | No, six languages only | Yes, wider language support | Textract |
Rates per 1,000 pages unless stated. Full rate card for the older service is on our AWS Textract pricing breakdown, and the two are compared feature by feature on Bedrock Data Automation vs Textract.
Put the three hyperscalers' structured extraction rates side by side and Bedrock Data Automation is the outlier twice over: it charges the highest base rate, and it is the only one that charges more when your schema gets wider.
| Service | The structured extraction meter | Per 1,000 pages | Charges more for more fields? | Volume discount |
|---|---|---|---|---|
| Amazon Bedrock Data Automation | Custom output blueprint | $40.00 | Yes, $0.50 per 1,000 per field above 30 | None |
| Azure AI Document Intelligence | Custom extraction | $30.00 | No, field count does not change the rate | Commitment tiers to $18.00 |
| Google Document AI | Custom Extractor | $30.00 | No, field count does not change the rate | $20.00 above 1M pages |
Every other major document AI service gets cheaper as you send more. Textract Detect Document Text drops from $1.50 to $0.60 per 1,000 pages above a million pages. Azure Document Intelligence sells commitment tiers that take Read down to $0.45. Google Document AI drops above five million pages. Bedrock Data Automation has no tier at all, so at one million pages a month, pulling plain text costs $1,500 on Textract and $10,000 on Bedrock Data Automation. If you are modelling a high-volume pipeline, that is not a rounding error, and it is the single strongest argument for keeping Textract underneath BDA and only calling the expensive service on the pages that actually need a schema.
Almost every guide repeats it as if it were the API ceiling. The sync API actually caps at 10 pages and 50 MB. The async API with the splitter enabled handles 3,000 pages and 500 MB, which is the highest single-document ceiling of any service we track.
Documents are supported in English, German, Spanish, French, Italian and Portuguese only. There is no Chinese, Japanese or Korean support, and AWS states plainly that vertical text alignment does not work. No rate table anywhere mentions this.
Textract, Azure and Google all get cheaper above a threshold. Bedrock Data Automation never does. The price on your first page is the price on your ten millionth, which inverts the usual advice that the hyperscaler wins at scale.
Word files are converted to PDF before processing, and AWS notes that page number mapping does not survive the conversion. If you need to cite a source page in a Word document, that trace is gone.
A blueprint holds up to 100 leaf fields, but the surcharge starts at 31. Those two numbers get conflated constantly. You can build a 100-field blueprint; you just pay $75.00 per 1,000 pages to run it.
Both printed and handwritten characters are read at the same rate, with no add-on charge, and any in-plane rotation is handled. Text has to be at least 15 pixels tall, which is about 8 point at 150 DPI.
The migrate-or-not decision, feature by feature, and where each one wins.
Read itEvery vendor normalized onto one unit, including the per-document and per-token ones.
Read itFull rate card, the 3-month free tier, and the models Textract does not have.
Read itMicrosoft's new service, which went the other way and undercut its own older one.
Read itFull rate card and the commitment tiers Microsoft does not publish.
Read itThe buyer pillar: which OCR API to actually pick, by workflow.
Read itBedrock Data Automation will hand you a JSON object shaped like the schema you asked for. What it will not do is tell you which of those fields the model was unsure about, route the failures to a person, enforce that the line items add up to the total, or push the result into your accounting system. You build all of that, and on most projects it costs far more than the $40.00 per 1,000 pages you were comparing.
DocuOCR prices at roughly $14 to $20 per 1,000 pages with the classification, the confidence scores, the human review screen and the exports already built. Whether that trade makes sense depends on how much of the pipeline you want to own.
No rate table tells you whether the fields come out right on the forms you actually process. Upload one and compare the output yourself.
Extract a document freeLast updated July 2026. Rates read from the official AWS Bedrock pricing page and re-verified before publication.
Z tej samej rodziny narzędzi
