🔍 Read the full analysis: What Can The Dharma Field Extractor Pull From Your Documents? 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 Dharma Field Extractor, a 2-billion-parameter model that converts document images into structured JSON, either by finding fields itself or following a user-provided schema. The company reports higher benchmark quality and lower estimated costs than four tested frontier model configurations, but full scores and independent validation are not provided.
Hugging Face has introduced Dharma Field Extractor, a 2-billion-parameter model that turns document images into structured JSON. The company says it outperformed four tested frontier model configurations on its benchmark and had substantially lower estimated operating costs, a claim that could matter to organizations processing large volumes of invoices and other records but has not been independently validated in the supplied material.
The model offers two request modes. In an open extraction request, it identifies information from a document’s labels and layout and returns key-value pairs. In a schema-directed request, a user names the fields and their types; the model is intended to return those keys, put null where a requested value is absent, and arrange table entries row by row.
Hugging Face says a single checkpoint supports multiple document types without per-type fine-tuning. Its examples include extracting suppliers, invoice numbers, due dates and totals from invoices, then sending requisition details to scheduling systems. The company presents the model as a way to move from page text to data that other software can use, rather than simply transcribing what appears on a page.
The system 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 training material included English documents from the public web, internally sourced Brazilian Portuguese documents and smaller sets in other European and Asian languages. Document categories included receipts, invoices, forms, bank statements, IDs and synthetic documents.
From Scanned Pages to Business Records
Document extraction addresses a recurring step in administrative work: turning information on pages into named fields that can be stored or passed to another system. OCR transcribes text, but does not by itself determine that a particular number is an invoice total or assign table entries to the right columns. A model that performs that mapping into a required JSON structure could reduce custom processing work if its output is reliable.
The two request modes serve different needs. Schema-directed output may fit systems that require fixed fields, while open extraction could help when a document’s structure is not known in advance. The reported cost advantage may also attract organizations processing many pages. But real-world value will depend on document quality, error rates, verification needs and integration costs, not just model inference estimates.
For now, the announcement is evidence of a defined product capability and a company-reported benchmark result, not proof that businesses can remove human review. Errors in financial or identity records can have consequences, so buyers would need to test the model against their own documents and workflows before relying on its output.
Products Worth Considering

Jackery Explorer 300 Portable Power Station, 292Wh

Jackery HomePower 3600 Plus Portable Power Station, 3584Wh, 3600W

LG UltraGear 34GX90SA-W 34-inch WQHD Curved OLED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
How the Model Builds on OCR
Hugging Face released DharmaOCR-LITE in April, according to the source material, for reading varied documents including contracts, court filings, invoices, IDs and forms. The year of that release is not specified. The Field Extractor adds a field-interpretation layer to OCR: instead of returning page text alone, it assigns values to fields and can organize table contents.
The company says its evaluation covered 546 documents and 1,638 cases, spanning open extraction, partial schemas and schemas that included fields absent from the source documents. It measured field-level F1 by path, including table position, combined with value similarity and the proportion of valid JSON outputs. Hugging Face reports a quality gap of 1.8 to 9.3 points against four frontier configurations, but the supplied account does not identify the score scale or list each model’s results.
Hugging Face also estimates costs roughly six to 147 times lower than those configurations. The estimates use average input and output tokens per page, including reasoning tokens, and providers’ published list prices. The Field Extractor’s own price is not stated. The source says the comparison models used low reasoning settings, with adaptive effort for Claude Opus 5.
““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 Results Still Need Validation
The supplied material does not include full benchmark tables, individual scores, the score scale or details sufficient to reproduce the evaluation. The reported cost figures are estimates rather than a stated price for the Field Extractor, and the source does not explain how costs would change with different document mixes, volumes or review requirements.
It is also unclear how the model performs on unseen documents, low-quality scans, and specific languages or document types. The level of human checking needed, the handling of sensitive documents, deployment requirements and availability of a public endpoint are not specified. Hugging Face says users can try the model, but the supplied information gives no release date for the announcement or further access details.
What Buyers and Evaluators Need
The next useful evidence would be complete benchmark results, reproducible evaluation details and independent tests on varied document sets. Organizations considering the model will also need to check how often it produces incorrect or missing fields, how it handles tables and absent values, and whether its output fits their existing software.
Until those details are available, the announcement supports a limited conclusion: Hugging Face has released a model designed to extract structured fields from document images, and the company reports favorable benchmark and cost comparisons. Whether that translates into dependable savings in a particular workflow remains to be established.
Key Questions
What does Dharma Field Extractor do?
It converts document images into structured JSON, either identifying fields on its own or filling fields specified by the user.
What kinds of documents can it process?
Hugging Face says its training examples included receipts, invoices, forms, bank statements, IDs and synthetic documents. That list does not establish how accurately it handles every document type in practical use.
How strong were the reported benchmark results?
Hugging Face reports a quality advantage of 1.8 to 9.3 points over four tested frontier configurations across 546 documents and 1,638 cases. The supplied material does not provide the score scale or individual results.
Is the model proven to cost less in a business workflow?
No. The company estimates costs six to 147 times lower using average token use and published list prices. The Field Extractor’s own price and the total cost of review, integration and error handling are not provided.
Can organizations use it without human review?
The supplied information does not establish that. Error rates and the amount of review needed in real workflows remain unclear, so organizations would need to test the model on their own documents.
Primary source: Hugging Face · via ThorstenMeyerAI.com
