Key points
- The model outputs typed JSON with nulls for missing fields, avoiding hallucination.
- It supports both open extraction and strict schema-directed requests.
Hugging Face published a blog post describing Dharma Field Extractor, a 2-billion-parameter model designed to pull specific data points from document images and return them as structured JSON.confirmed The company claims this specialized approach beats general-purpose frontier large language models on this task while costing significantly less to run.company claimThe model turns pictures of documents into organized data that computers can easily read and store.
Evidence: confirmed
Dharma Field Extractor: Any Document In, Structured Fields in JSON Outaifleet-huggingface
The claim
The post states that the model handles any document type without per-type fine-tuning. It can extract fields from invoices, IDs, medical requisitions, and contracts. Hugging Face asserts that the model returns exactly the fields requested, with correct typing and nesting, matching the structure of downstream systems.
The response
The material does not include independent verification or third-party testing results. The performance comparisons are based on tests conducted by the makers. The blog post contrasts the model against the high latency and pricing of frontier APIs, positioning the new tool as a more efficient solution for high-volume business tasks.
What the evidence means
The evidence rests on the publisher's own benchmarks. The post argues that general OCR tools return unstructured text, requiring additional coding to map values to fields, while general LLMs are expensive for this narrow use case. The Field Extractor aims to bridge that gap by providing ready-to-use JSON.
In practice
Users can send a document image with either an open request or a specific schema. In open mode, the model identifies structured fields from context and layout, returning key-value pairs. In schema-directed mode, users define exact field names and types (string, number, boolean). The model returns only those keys, filling tables row by row. If a requested field is absent, it returns null rather than inventing a value. This allows direct integration into existing databases without extra mapping code.
Why it matters
- The Field Extractor outperforms frontier LLMs on document field extraction at a fraction of the cost and latency.
- It returns structured JSON with typed fields and null values for missing data, eliminating the need for custom code to map OCR text to database columns.
- The model supports both open extraction for unknown document structures and schema-directed extraction for strict system compatibility.
- A single checkpoint handles all document types without per-type fine-tuning, reducing maintenance overhead for high-volume processing.
Topics: models, document processing, Hugging Face, JSON, OCR
Updated
Then / now / next
Businesses used general-purpose LLMs or custom code to extract fields from documents.
A specialized 2B model offers schema-directed JSON extraction at lower cost.
Watch for independent benchmarks comparing this model against other specialized OCR tools.
Story timeline
- Then Businesses used general-purpose LLMs or custom code to extract fields from documents.
- 04 Nov 2026 Dharma Field Extractor: Any Document In, Structured Fields in JSON Out
- Now A specialized 2B model offers schema-directed JSON extraction at lower cost.
- Next Watch for independent benchmarks comparing this model against other specialized OCR tools.
The strongest doubt
Counterpoint from the sources
The performance claims are self-reported by Hugging Face and lack independent third-party validation in the provided material.
Ask the article
Local keyword search of this article; no generated answers.Which of these is confirmed?
Check the answer
Hugging Face published a blog post describing Dharma Field Extractor, a 2-billion-parameter model designed to pull specific data points from document images and return them as structured JSON.
The matched primary source and announcement wording support the confirmed label.
Did this article give you the evidence you needed?
Your choice stays on this device. No public tally.Questions readers ask
Does the model require fine-tuning for different document types?
No, the blog post states that the same checkpoint handles all document types without per-type fine-tuning.
What happens if a requested field is not in the document?
The model returns null for that field instead of inventing a value.
Can the model extract tables from documents?
Yes, when asked for a table, it fills it row by row and column by column, matching cells to the correct keys.
Connected stories
Reading receipt
- The model outputs typed JSON with nulls for missing fields, avoiding hallucination.
- It supports both open extraction and strict schema-directed requests.
Still open: How does the model perform on handwritten or low-quality scans?
Still open: What is the exact latency compared to frontier models?