Dharma-AI's Gabriel Pimenta ships a 2B model that extracts document fields as JSON
Co-founded by Gabriel Pimenta, Dharma-AI built its 2B extractor on its own OCR model and trained it on 43,384 field-level examples.
By Ryan Merket · Published
Primary source: Hugging Face Newsroom
Why it matters
Document extraction is a high-volume, measurable task where per-page frontier API costs compound fast. A self-hosted 2B specialist that holds its own on field-level accuracy changes the buying calculation for back-office pipelines, and tests whether Dharma-AI's specialization thesis extends from OCR to structured output.
Dharma-AI co-founder Gabriel Pimenta de Freitas Cardoso and three colleagues published the Dharma Field Extractor on November 4th, 2026, a 2-billion-parameter model that takes a document image and returns fields a downstream system asks for, typed as JSON. The announcement on Hugging Face lists Paulo Henrique de Medeiros Araujo, Bruno Duarte and Francisco de Almeida Rocha Alves alongside Cardoso as contributors.…
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.