The Agent RaceJune 23, 2026via MarkTechPost
Datalab Releases lift: A 9B Open-Weights Vision Model That Extracts Structured JSON From PDFs Using Schemas
Why it matters
A practical, production-ready vision model for document automation that solves a real enterprise pain point (PDF-to-structured-data) with schema constraints and trained abstention. Open-weights release makes it immediately deployable.
Key signals
- Model size: 9B parameters
- Open-weights release
- 90.2% field accuracy on 225-document benchmark
- Schema-constrained decoding for output validity
- Trained abstention to avoid hallucinations on missing fields
- Converts PDFs and images to schema-matching JSON
- Released by Datalab
- Published June 23, 2026
The hook
90.2% field accuracy. Datalab's 9B vision model turns PDFs into valid JSON without hallucination—and it's open-weights.
Datalab released lift, a 9B open-weights vision model that turns PDFs and images into schema-matching JSON. It uses schema-constrained decoding for valid structure and trained abstention to return null instead of hallucinating absent fields, scoring 90.2% field accuracy on a 225-document benchmark.