Saved in:
Bibliographic Details
Main Authors: Taghadouini, Said, Cavaillès, Adrien, Aubertin, Baptiste
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2601.14251
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912835920986112
author Taghadouini, Said
Cavaillès, Adrien
Aubertin, Baptiste
author_facet Taghadouini, Said
Cavaillès, Adrien
Aubertin, Baptiste
contents We present \textbf{LightOnOCR-2-1B}, a 1B-parameter end-to-end multilingual vision--language model that converts document images (e.g., PDFs) into clean, naturally ordered text without brittle OCR pipelines. Trained on a large-scale, high-quality distillation mix with strong coverage of scans, French documents, and scientific PDFs, LightOnOCR-2 achieves state-of-the-art results on OlmOCR-Bench while being 9$\times$ smaller and substantially faster than prior best-performing models. We further extend the output format to predict normalized bounding boxes for embedded images, introducing localization during pretraining via a resume strategy and refining it with RLVR using IoU-based rewards. Finally, we improve robustness with checkpoint averaging and task-arithmetic merging. We release model checkpoints under Apache 2.0, and publicly release the dataset and \textbf{LightOnOCR-bbox-bench} evaluation under their respective licenses.
format Preprint
id arxiv_https___arxiv_org_abs_2601_14251
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle LightOnOCR: A 1B End-to-End Multilingual Vision-Language Model for State-of-the-Art OCR
Taghadouini, Said
Cavaillès, Adrien
Aubertin, Baptiste
Computer Vision and Pattern Recognition
We present \textbf{LightOnOCR-2-1B}, a 1B-parameter end-to-end multilingual vision--language model that converts document images (e.g., PDFs) into clean, naturally ordered text without brittle OCR pipelines. Trained on a large-scale, high-quality distillation mix with strong coverage of scans, French documents, and scientific PDFs, LightOnOCR-2 achieves state-of-the-art results on OlmOCR-Bench while being 9$\times$ smaller and substantially faster than prior best-performing models. We further extend the output format to predict normalized bounding boxes for embedded images, introducing localization during pretraining via a resume strategy and refining it with RLVR using IoU-based rewards. Finally, we improve robustness with checkpoint averaging and task-arithmetic merging. We release model checkpoints under Apache 2.0, and publicly release the dataset and \textbf{LightOnOCR-bbox-bench} evaluation under their respective licenses.
title LightOnOCR: A 1B End-to-End Multilingual Vision-Language Model for State-of-the-Art OCR
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2601.14251