LightOn releases LightOnOCR-3, open OCR models that also map document layout
LightOn published LightOnOCR-3 in 0.8B, 1B and 4B sizes. Besides transcribing text, the models output labeled bounding boxes for document regions, describe images and pull numbers from charts.
French AI company LightOn released the LightOnOCR-3 family on Hugging Face on October 8, in three sizes: 0.8B, 1B and 4B parameters.
LightOn says the new generation is faster and more accurate than its predecessor and adds visual understanding: the models can return labeled bounding-box coordinates for every region of a document, write image descriptions and extract numerical data from figures and charts. The company pitches this as a simpler alternative to multi-step document pipelines.
The models are released under the Apache 2.0 license for research and commercial use, with a demo Space on the Hub.