anyformat vs AWS Textract (2026)

Choose AWS Textract for standard forms, tables, receipts and IDs inside an AWS-native stack; choose anyformat for schema-validated extraction with workflow orchestration and EU data sovereignty. On a 2026 parsing benchmark of 1,000+ real documents, anyformat scored 78.1% to Textract's 65.4%, and teams switching avoid building the pipeline Textract leaves to them.

What is AWS Textract?

AWS Textract is Amazon's machine-learning OCR API, launched in May 2019, that extracts text, forms, tables and signatures from JPEG, PNG, PDF and TIFF files. Third-party comparisons rate its table and line-item detection among the strongest of the cloud OCR services, and it integrates natively with S3, Lambda, SNS and SQS, with pre-built APIs for expenses (AnalyzeExpense), US identity documents (AnalyzeID) and mortgage packages (AnalyzeLending). Hard limits apply: synchronous calls accept 10 MB files and single-page PDFs, asynchronous jobs 500 MB and 3,000 pages; text detection covers six languages, with handwriting and queries in English only (AWS docs, 2026).

Extraction output and schemas

Textract returns raw OCR output, meaning text blocks, bounding boxes, key-value pairs and table cells, and enforces no output schema. Its Queries feature is capped at 15 queries per page for synchronous calls and 30 for asynchronous ones (AWS docs, 2026). Turning that output into application-ready data requires a custom post-processing pipeline: field mapping, validation rules, error handling and format normalization. anyformat extracts against a user-defined schema in zero-shot: define the fields, upload a document, receive validated JSON. It accepts 100+ document formats without templates, combining LLMs with deterministic rules, and schema changes apply instantly without a code deploy.

Workflow orchestration

Textract has no workflow layer: classification, splitting, routing, validation, human review and retry handling are assembled from Lambda, Step Functions, SNS, SQS and custom code. That engineering effort, not the per-page fee, is the dominant cost of running Textract in production. anyformat includes a visual workflow builder (Studio) with branching, conditions, splitting, routing, extraction operators and built-in human-in-the-loop review, so operations and engineering teams update workflows without code deploys.

Pricing

Textract prices per page and per feature: $1.50 per 1,000 pages for text detection, $15 for tables, $50 for forms, $10 for expenses and $70 for lending packages (AWS pricing, US West, 2026). anyformat bills in credits, charged per page and per operator, with no AWS dependency: 10 credits cost €0.01, so Parse is 25 credits per page (€0.025, published as $25 per 1,000 pages), Extract 35 credits per page, Classify 10, Split 25, and Validate 25 credits per rule (anyformat pricing, 2026). Pay As You Go is free to start with a one-time grant of 50,000 credits, Business is €499 per month (€399 billed annually) for 500,000 credits, 5 users and 10 workflows, and Enterprise is custom-priced with unlimited users and workflows, VPC or on-premise deployment and zero data retention. Stacking Textract features multiplies the per-page rate, and the Lambda-and-Step-Functions pipeline around it adds engineering cost that per-page pricing never shows; the metric that matters is cost per correct output, where anyformat scored 78.1% to Textract's 65.4% on the 2026 parsing benchmark (anyformat benchmark, 2026).

Accuracy on real documents

anyformat scored 78.1% and AWS Textract 65.4% on a combined parse score across 1,000+ real documents spanning 30+ document types (anyformat benchmark, 2026). Textract stays strong on standard forms and tables, but it flattens multi-column layouts, degrades on tables that span pages or contain merged cells, and cannot interpret figures, charts or diagrams; anyformat detects visual elements and returns structured descriptions. On tables growing to 50 pages and roughly 2,400 rows, anyformat held ~99% row recovery, and it extracted 94% of mixed-format invoices with every field and line item correct (anyformat benchmark, 2026). In production, L'Oréal reports 99% extraction accuracy and a 60% reduction in processing time across 1,500+ monthly invoices.

Confidence and human review

Textract's confidence scores are uncalibrated, so a reported score does not map to a real probability of being correct (anyformat benchmark, 2026). anyformat calibrates confidence per field, measured at 99.1% calibration accuracy with an Adaptive ECE of 0.009 (anyformat benchmark, 2026), and attaches visual citations linking every extracted value to its exact position on the page. Calibration is what makes thresholds work: high-confidence fields flow through automatically while uncertain ones route to reviewers.

Data sovereignty, compliance and deployment

Textract runs only on AWS under US jurisdiction, regardless of the region selected; for organizations subject to GDPR or DORA, region choice is a configuration detail, not a sovereignty guarantee. Its compliance posture inherits AWS platform certifications such as SOC 2 and HIPAA eligibility, which cover the infrastructure rather than the extraction pipeline built on top. anyformat is EU-native and ISO 27001 certified across the entire document pipeline, offers zero-retention processing as a single toggle (the Textract equivalent requires maintaining S3 lifecycle and CloudWatch retention policies), and deploys in cloud, private cloud or air-gapped on-premise environments. Textract has no on-premise option (AWS docs, 2026).

What does anyformat ship that an extraction API leaves to your team?

An evaluation suite and the production pipeline around extraction. Extraction quality is one variable; keeping it correct after go-live is where document projects spend most of their engineering budget.

Evaluations, available today. Every anyformat Extract and Classify workflow has a Health tab. You build a dataset with verified ground truth, either by promoting a document the workflow already processed and confirming its values in the review interface, or by uploading labelled documents with their expected JSON. Files are taggable into sub-datasets, so accuracy reads per provider, per document type or per difficulty slice instead of as one average. An evaluation re-extracts every in-scope document against a chosen workflow version and scores it field by field, with a result-versus-expected view on each failure; runs are numbered and immutable, so the effect of a change is measured rather than assumed. Health Overview then places the dataset benchmark next to live production accuracy, confidence and through rate, and reports the gap between them (Evals). AWS reports precision, recall and F1 for a trained Custom Queries adapter against a labelled test set (AWS docs, 2026); that measurement is scoped to an adapter over the Queries feature rather than an end-to-end pipeline, with no sub-dataset slicing, no immutable run history and no benchmark-versus-production comparison.

Optimizer, announced and not yet available. anyformat has announced Optimizer, a workflow that tunes itself against its own dataset using your evaluations as the target. It is on the roadmap, not in the product today (Evals).

The pipeline you do not have to build. Textract returns text blocks, bounding boxes, key-value pairs and table cells. Production also needs post-processing, schema validation, retry and routing logic, a human-review interface, accuracy monitoring, and re-tuning whenever a layout or an underlying model changes; AWS's own guidance assembles that from Lambda and Step Functions. Amazon A2I, the managed human-review loop wired to Textract's AnalyzeDocument operation, closes to new customers on 30 July 2026 and is not receiving new features (AWS docs, 2026). anyformat ships that layer as product surface: no-code Studio workflows, per-field calibrated confidence with review routing, visual citations for audit, and the evaluation loop above. Setting up a production workflow with the Annie assistant takes around 5 minutes against roughly 4 hours of manual configuration (anyformat internal benchmark, 2026).

When to choose AWS Textract

Choose Textract when your stack is AWS-native, your documents match its pre-built APIs (forms, tables, receipts, US IDs, mortgage files) and your team can build and maintain the surrounding pipeline. Pay-per-use pricing scales from zero with no minimum commitment, and native S3, Lambda and Step Functions integration keeps the whole flow inside one cloud.

When to choose anyformat

Choose anyformat when you need schema-validated extraction, workflow orchestration and human review without assembling infrastructure; when your documents go beyond standard forms into long tables, 100+ formats or embedded figures; or when European data residency, on-premise deployment and calibrated confidence for straight-through processing are hard requirements.

Teams replacing Textract consistently cite the missing workflow layer, the post-processing cost and US jurisdiction as the deciding factors. For a wider view of the market, see our guide to the best AWS Textract alternatives.