Changelog for the Data Extraction API

Improves table extraction for dropped or split tables, wrapped rows, stacked records, shaded regions, merged cells, and false table detection.

Improves scanned-document accuracy by running optical character recognition (OCR) alongside vision model analysis and resolving disagreements with evidence fusion. (L#NAVI-124, L#NAVI-125, L#NAVI-130)

Improves OCR language handling so Latin scans are less likely to be misclassified as Cyrillic, Arabic, or Greek, and more configured OCR languages are honored. (L#NAVI-148, L#NAPY-24)

Charts now return their data as tables by default, and QR codes and barcodes now return their decoded content.

Increases extract schema limits: schema size from 32 KB to 128 KB, total fields from 500 to 2,000, properties per object from 50 to 200, required entries from 50 to 200, and description length from 1,024 to 4,096 characters. (#58687)

Clarifies the classify API documentation and specification: Scores are independent, and non-zero text/image weights are ignored by the engine. (#58039)

Updates parse to return an actionable 400 error for password-protected documents instead of a generic 500. (#57421)

Fixes classify reporting placeholder pagesProcessed values, and fixes billing per request instead of per processed page. (#56853)

Updates extract billing so requests are billed as parse and extract components. Responses now include a price_composition breakdown. (#53930)

Adds date-based Data Extraction API versioning via the x-nutrient-api-version header. The initial API version is 2026-05-25. (#53345)

Adds the OpenAPI schema for the Data Extraction API. (#52250)

Introduces the initial Data Extraction pipeline with extraction support for PDFs, images, and Office documents. (#51564)