🔍 Read the full analysis: A Practical Guide To The Dharma Field Extractor on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Hugging Face has introduced the Dharma Field Extractor, a 2-billion-parameter model that turns document images into JSON, either by finding fields itself or filling a user-provided schema. The company reports higher benchmark quality and substantially lower estimated costs than four tested frontier model configurations, but full scores, pricing and independent validation were not provided in the supplied material.
Hugging Face has introduced the Dharma Field Extractor, a 2-billion-parameter model designed to turn document images into structured JSON, as described in the original analysis. The company says it performed better than four tested frontier model configurations on its benchmark, at substantially lower estimated cost, but the supplied information does not include full scores or independent validation.
The model has two modes. In open extraction, it identifies information available from document labels and layout and returns key-value pairs. In schema-directed extraction, a user specifies field names and types; the model is intended to return those keys, place null where a requested value is absent, and arrange table entries row by row.
Hugging Face says a single checkpoint handles different document types without fine-tuning for each one. Its examples include extracting supplier names, invoice numbers, due dates and totals, then passing requisition details to scheduling software. The company describes the model as a step beyond OCR: OCR transcribes page content, while field extraction assigns information to usable fields.
The extractor builds on DharmaOCR-LITE v1.1, a 2-billion-parameter OCR model, with a LoRA adapter trained on 43,384 examples, according to Hugging Face. The examples covered English documents from the public web, internally sourced Brazilian Portuguese documents and a smaller selection of other European and Asian languages. Document types included receipts, invoices, forms, bank statements, IDs and synthetic documents.
From Scans to Structured Records
Document extraction addresses a recurring handoff in data processing: turning content on a page into records that business software can use. If a model reliably fills required fields and formats them correctly, organizations could reduce custom mapping and manual entry for invoices, forms and other records. Schema-directed JSON is relevant to systems that expect fixed keys; open extraction may help when fields vary or are not known in advance.
The reported cost advantage could matter for organizations handling large volumes of pages. However, the company’s estimates use measured average token consumption and providers’ published list prices. Actual workflow costs may also depend on document quality, processing volume, error handling and human review. The supplied benchmark summary does not establish the model’s performance in any particular organization’s operations.
Products Worth Considering
Bose SoundLink Flex Bluetooth Speaker (2nd Gen) – Portable Outdoor Speaker with Hi‑Fi Audio, Waterproof and Dustproof, USB‑C, Up to 12 Hours
EcoFlow Delta 3 Portable Power Station – 1024 Wh, 1800 W
Anker SOLIX C300 Portable Power Station, 288Wh
As an affiliate, we earn on qualifying purchases.
Built on DharmaOCR-LITE
Hugging Face says it released DharmaOCR-LITE in April for reading varied documents, including contracts, court filings, invoices, IDs and forms. The supplied material does not specify the year of that release. The Field Extractor extends that OCR work by adding a layer that assigns recognized content to named fields and, for tables, relates entries to rows and columns.
The company says its benchmark covered 546 documents and 1,638 cases, including open extraction, partial schemas and schemas with fields that did not exist in the document. It evaluated field-level F1 by path, including table position, alongside value similarity and the share of outputs that were valid JSON. Hugging Face reports a quality gap of 1.8 to 9.3 points over four frontier configurations, but the supplied summary does not identify the score scale or provide each model’s results.
“The Field Extractor takes a document image and either a request for everything it can find, or an exact schema to fill, and returns structured JSON.”
— Hugging Face
Benchmark Details Still Missing
The company also reports estimated costs roughly six to 147 times lower than those of the tested configurations. The estimates use average input and output tokens per page, including reasoning tokens, and providers’ published list prices. The supplied material does not state the Field Extractor’s own price, provide full benchmark tables, or explain the individual cost comparisons in detail.
Independent reproduction and results on unseen document sets have not been supplied. It is also unclear how error rates vary by language and document type, how the model performs on poor-quality scans, and how much human review is needed. The benchmark included multiple document categories, but its summary does not show whether those results predict performance on a specific organization’s records. Details such as deployment requirements and a public endpoint are also absent from the material.
More Testing Needed in Workflows
Hugging Face says users can try the model, but the supplied information does not give a release date or details of an endpoint or deployment setup. The next evidence that would clarify its practical value includes complete benchmark results, reproducible evaluation methods and tests by organizations using varied documents and field requirements.
Teams considering the model would also need to decide how to handle incorrect or missing values, sensitive documents and cases requiring human checks. Until those questions are tested against real workflows, the reported benchmark and cost estimates indicate potential rather than proven savings or reliability across business settings.
Key Questions
What does the Dharma Field Extractor do?
It processes document images and returns structured JSON, either by identifying fields itself or by filling a schema supplied by the user.
How is it different from OCR?
OCR transcribes text from a page. The Field Extractor is intended to go further by assigning information to named fields and organizing table entries into structured records.
What benchmark results did Hugging Face report?
Hugging Face says the model scored 1.8 to 9.3 points higher than four tested frontier configurations across a benchmark of 546 documents and 1,638 cases. The supplied summary does not include the score scale or individual results.
Is the model cheaper than the tested alternatives?
Hugging Face estimates costs roughly six to 147 times lower for the tested configurations, using average token use and published list prices. The Field Extractor’s own price and the full calculation are not provided in the supplied material.
Has its performance been independently verified?
No independent validation is described in the supplied material. Performance on unseen documents and the amount of human review required remain unclear.
Primary source: Hugging Face · via ThorstenMeyerAI.com
