🔍 Read the full analysis: Dharma Field Extractor For AI-Powered Document Processing 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 structured JSON, either by finding fields itself or filling a user-defined schema. The company reports higher benchmark quality and lower estimated costs than four tested frontier model configurations, but the supplied information omits full scores, the extractor’s own price and independent validation.
Hugging Face has introduced the Dharma Field Extractor, a 2-billion-parameter model that converts document images into structured JSON. The company says it can identify fields without a predefined list or fill a supplied schema, and reports that it outperformed four tested frontier model configurations on its benchmark at substantially lower estimated cost.
The extractor offers two modes. In an open request, it identifies information from document labels and layout and returns key-value pairs. In a schema-directed request, a user specifies field names and types; the model is intended to return those keys, use null when a requested value is absent, and organize tables row by row. Hugging Face says one checkpoint handles different document types without per-type fine-tuning.
The company describes uses such as pulling suppliers, invoice numbers, due dates and totals from invoices, then passing requisition details to scheduling systems. It says the benchmark included 546 documents and 1,638 cases, covering open extraction, partial schemas and schemas with nonexistent fields. Hugging Face reports a quality gap of 1.8 to 9.3 points against the four tested configurations and costs roughly six to 147 times lower. The supplied material does not provide the individual scores or identify the score scale.
The reported cost estimates use average input and output tokens per page, including reasoning tokens, and providers’ published list prices. The source does not state the Field Extractor’s own price. It also does not provide enough information to independently reproduce the benchmark or establish how the results translate to a particular organization’s documents.
From Scanned Pages to Business Records
Document processing often involves more than reading words off a page: businesses need those words assigned to fields that software can store and act on. A model that reliably returns valid, schema-shaped JSON could reduce custom mapping and manual data entry in recurring workflows such as invoice handling. Open extraction may also help where the relevant fields are not known in advance.
The cost claim may draw interest from organizations handling large document volumes, but the reported estimate is not the same as the total cost of a deployed workflow. Real-world costs can depend on document quality, volume, model errors, human review and how extracted data is checked before it reaches other systems. The announcement provides a benchmark comparison, not evidence that the model will perform equally well across business settings.
Products Worth Considering
Jackery Explorer 300 Portable Power Station, 292Wh
Bose SoundLink Flex Bluetooth Speaker (2nd Gen) – Portable Outdoor Speaker with Hi‑Fi Audio, Waterproof and Dustproof, USB‑C, Up to 12 Hours
Anker SOLIX C300 Portable Power Station, 288Wh
As an affiliate, we earn on qualifying purchases.
Built on DharmaOCR-LITE
The Field Extractor builds on DharmaOCR-LITE v1.1, a 2-billion-parameter model specialized for optical character recognition. OCR transcribes page content; field extraction adds an interpretation step by assigning values to named fields and organizing items such as table entries. Hugging Face describes the new model as a task-specific extension of its document-reading work.
The company says it trained a LoRA adapter on 43,384 examples. Those examples included English documents from the public web, internally sourced Brazilian documents in Portuguese and a smaller set of other European and Asian languages. The described categories included receipts, invoices, forms, bank statements, identification documents and synthetic documents. Hugging Face previously released DharmaOCR-LITE in April, according to the supplied material, but the year is not specified.
““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 Limits and Deployment Questions
The reported benchmark and cost comparisons remain company-reported results. The supplied material does not include full score tables, individual model results, the score scale, the extractor’s price or an independent replication. It says the comparison models used low reasoning settings, with adaptive effort for Claude Opus 5, but does not give enough detail to assess how those settings affected the results.
Performance on unseen documents, error rates by language and document type, and the amount of human review needed are also unclear. The material does not specify deployment requirements or identify a public endpoint. It says the model can be tried, but gives no release date. Teams would also need to decide how to handle incorrect values, sensitive documents and fields that are missing from the source page.
Independent Testing Will Matter
The next useful evidence would include complete benchmark tables and reproducible evaluation details, followed by testing on varied documents outside the company’s evaluation set. Organizations considering the model will need to compare its output with their own requirements, including document quality, languages, table formats and tolerance for errors.
Before relying on extracted values in operational systems, teams will need a process for checking uncertain or incorrect results and protecting sensitive documents. The supplied material does not announce a specific next benchmark, deployment milestone or pricing update, so the timing of further details is unknown.
Key Questions
What does the Dharma Field Extractor do?
It processes document images and returns structured JSON, either by identifying fields on its own or by filling fields defined in a user-supplied schema.
How did Hugging Face say it performed?
Hugging Face reports a quality advantage of 1.8 to 9.3 points over four tested frontier model configurations on a benchmark of 546 documents and 1,638 cases. The supplied information does not include the full scores or score scale, and the results have not been independently established here.
How much does the model cost?
The announcement reports estimated costs roughly six to 147 times lower than those of the tested configurations, based on token use per page and published list prices. It does not state the Field Extractor’s own price, so those estimates should not be treated as a confirmed charge for using it.
Which documents and languages were included in training?
Hugging Face says training included receipts, invoices, forms, bank statements, identification documents and synthetic documents. The examples covered English, internally sourced Brazilian documents in Portuguese and a smaller set of other European and Asian languages.
Is the model proven to work reliably in business workflows?
The supplied material does not establish that. Independent testing, performance on unseen documents, error rates and the required level of human review remain unclear.
Primary source: Hugging Face · via ThorstenMeyerAI.com
