SPARK Prism
A Software Spark-developed multimodal model line for document, image, and visual inspection workflows with structured outputs.
VERSION0.4FAMILYVision and document understandingCONTEXTUp to 128K target; confirmed per deploymentRELEASEControlled preview
- INPUTS
- Images, documents, text instructions
- OUTPUTS
- Text, JSON schema, regions and classifications
- OWNERSHIP
- Software Spark-developed model program; artifact access subject to evaluation agreement.
- Document classification and field extraction
- Visual inspection with reviewer escalation
- Image-grounded search and summarization
- Multimodal workflow routing
- Small visual details and poor scans can reduce accuracy
- Spatial reasoning requires task-specific evaluation
- Not intended for biometric identification
- Sensitive document use requires an approved data boundary
- Field-level extraction precision and recall
- Document and image classification accuracy
- Structured-output conformance
- Reviewer escalation and uncertainty behavior
No external benchmark score is claimed on this card. Reproducible results will follow the SPARK benchmark protocol.