AI BriefingDharma Field Extractor Outperforms Frontier LLMs on Document Field Extraction
·2026.11.05 02:50
Key point
The 2B parameter model achieves a 0.834 field-level F1 score, surpassing Claude Opus 5 by 1.8 points and GPT-5.6 Luna by 9.3 points.
Details
- Dharma-AI released Dharma Field Extractor, a 2B parameter model that extracts specific fields from document images into structured JSON.
- The model outperforms general-purpose frontier LLMs, scoring 0.834 on a 1,638-case benchmark compared to 0.816 for Claude Opus 5 and 0.741 for GPT-5.6 Luna.
- It supports two modes: open extraction without a schema and schema-directed extraction with specific field types, returning null for missing fields.
- Self-hosting costs are significantly lower than API pricing, ranging from 6x to 147x cheaper than frontier models based on a $2.24/hour L40S GPU rate.
- Trained on 43,384 examples using LoRA adapters on the DharmaOCR-LITE v1.1 base, the model generalizes well to unseen datasets like SROIE and FUNSD.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.