Announcing a 2-billion-parameter document extraction model… self-evaluation score of 0.834

Dharma-AI released Dharma Field Extractor, a 2-billion-parameter model that pulls specified fields from document images—such as vendor names, invoice numbers, due dates, and totals—and returns them as structured JSON. The company reported a self-evaluated score of 0.834 across 1,638 cases, with target use cases including invoices, forms, ID cards, and bank statements. The model can either locate values on its own when target fields are not predefined or extract fields specified in advance. The release, announced on Hugging Face, matters because it targets structured data capture rather than full-text OCR, a common bottleneck in document-heavy finance workflows. What to watch next is independent benchmarking of the 0.834 self-reported score and adoption across invoice and statement-processing pipelines.