Dharma-AI Releases Field Extractor 2B for Document-to-JSON Extraction
The 2B model returns schema-directed JSON from document images, with missing fields returned as null; weights and licensing terms are not yet published.
Dharma-AI announced Field Extractor 2B on 4 November 2026, a document-understanding model designed to convert document images into structured JSON output. The model is built on DharmaOCR-LITE v1.1, a 2-billion-parameter base released in April 2026, combined with a LoRA adapter trained on 43,384 examples. The company positions the release as part of a broader shift in document processing, in which systems return programmable data contracts rather than page-faithful text.
Field Extractor operates in two modes. In open mode, the model detects field structure from the document layout, including labelled values and header-bearing tables. In schema-directed mode, the client defines typed fields using its own key names. Missing fields are returned as null values instead of generated content, an approach the company says addresses hallucination risks in key information extraction.
The vendor claims the model outperforms the frontier LLMs it tested on this task at lower cost and latency; these figures are self-reported and not independently verified. Model weights and licensing terms have not yet been published.