Public KG record
Dharma Field Extractor
Recorded overview
A 2-billion-parameter model that takes a document image and a request or schema,
An evidence-backed entity from the project knowledge graph; detailed tool metadata is still being added.
Recorded details
- Benchmarks: scored 0.834 on full benchmark covering 546 documents across three request scenarios
- Capabilities: Open field extraction without a schema
- Capabilities: Schema-directed field extraction matching exact keys and types
- Capabilities: extracts exactly the fields you ask for, from any document type, outperforming every frontier LLM we tested, at a fraction of the cost
- Deployment: runs self-hosted
- Languages: English and Portuguese, with a smaller share covering other European and Asian languages
- Supported types: input document image, output structured JSON
- Pricing: GPU time on a single L40S at $2.24 per hour
Evidence coverage
The KG currently records 13 known information dimensions and has 5 undocumented dimensions, linked to 1 public source articles. These counts describe record coverage, not model performance or verified fit.
Related public records
- DharmaOCR-LITE v1.1 — based on
- Claude Opus 5 — outperforms
- GPT-5.6 Luna — outperforms
- GPT-5.6 Sol — outperforms
- LoRA — uses