Unstructured Launches New Transform Document Parsing Models

In a press release, Unstructured launched new models for Transform, its document parsing service for PDFs, forms and scans. The models are available now with 10,000 free pages.

In company testing across 224 business document pages, Transform produced fewer table errors and less invented content than GPT-6 Astra, Claude Opus 5.5 and Gemini 3.8 Flash. Unstructured says the other models produced up to 2.3 times as many table errors and about 50% more content that was absent from the source documents.

Transform returns tables as grids, forms as label and value pairs, and text extracted from scanned PDFs. Results include element types, page numbers and coordinates. The service costs $0.015 per page and offers Balanced and Best parsing options, with dedicated deployments available on AWS, Azure and Google Cloud.

The comparison is Unstructured's own, run on SCORE-Bench, a benchmark the company built, and Unstructured sells the Transform service it tested.