Datalab Releases lift: A 9B Open-Weights Vision Model That Extracts Structured JSON From PDFs Using Schemas

Datalab has introduced lift, a groundbreaking 9-billion parameter open-weights vision model designed to automatically convert PDF documents and images into structured JSON data that conforms to user-defined schemas. This advancement addresses a critical pain point in data processing workflows where extracting and organizing information from unstructured documents has traditionally required manual intervention or expensive proprietary solutions.

The lift model represents a significant leap forward in document intelligence technology. Operating at 9 billion parameters, lift utilizes schema-constrained decoding to ensure that all generated output matches specified data structures with precision. A key innovation embedded in the model is its trained abstention capability—rather than generating false information when data is unavailable, lift intelligently returns null values, substantially reducing hallucination errors that plague many AI systems processing documents.

According to Datalab's evaluation on a 225-document benchmark, lift achieved 90.2% field accuracy, demonstrating its reliability for enterprise-grade applications. The model's open-weights availability means organizations can deploy, modify, and integrate it directly into their infrastructure without proprietary restrictions.

• Cost reduction: Eliminates expensive document processing services by enabling in-house deployment of a capable vision model
• Data accuracy: Schema-constrained decoding and abstention training minimize hallucinations and invalid outputs
• Accessibility: Open-weights distribution democratizes advanced document intelligence for companies of all sizes
• Integration flexibility: JSON output format enables seamless integration with existing data pipelines and enterprise systems
• Scalability: Organizations retain full control over model deployment and can optimize for their specific use cases

The release of lift addresses a fundamental challenge in enterprise data management: converting paper and digital documents into machine-readable structured data at scale. Traditional approaches require expensive human review or proprietary AI services with limited customization. By offering an open-source alternative with strong accuracy metrics and built-in safeguards against hallucination, Datalab enables organizations to automate document processing workflows while maintaining data integrity. This development could significantly reduce operational costs while improving efficiency across industries reliant on document-heavy processes, from financial services to healthcare and legal sectors.