LightOn Introduces LightOnOCR-3 for Advanced Document Intelligence

LightOn has announced the release of LightOnOCR-3, a family of AI models designed to enhance document intelligence. These models deliver top-tier performance in French-language document and handwriting recognition, surpassing US and Chinese counterparts. Building on the capabilities of its predecessor, LightOnOCR-3 excels not only in text recognition but also in image description, data extraction from charts, and preserving document structures. Available in 0.8-billion and 4-billion parameters, these models offer cost-effective solutions with fast processing times.

On OlmOCR-Bench, LightOnOCR-3 ranks second, while on ParseBench, it leads in document layout and image interpretation. Its performance is particularly strong with French-language documents, providing significant advancements in handwriting recognition. It addresses needs in finance, administration, and technical document search by processing comprehensive document information including text, images, and structured data in one step.

Organizations benefit from faster and cheaper document processing as LightOnOCR-3 models can reduce processing costs to under a cent per thousand pages when deployed on their infrastructure. Released under the Apache 2.0 license, these models support commercial use, modification, and redistribution, fitting diverse needs.