LightOnOCR-2-1B: a lightweight high-performance end-to-end OCR model family

Back to Articles

Overview Links Capabilities Key benchmarks Transcription quality Speed What we’re releasing Default model: best OCR OCR + lightweight layout cues: bbox-capable variants Base checkpoints (for fine-tuning / research) Acknowlegments

Overview

We’re releasing LightOnOCR-2-1B, our second-generation 1B-parameter, end-to-end vision-language OCR model optimized for state-of-the-art conversion of document pages (PDF renders) into clean, naturally ordered text without relying on multi-stage pipelines. Alongside transcription, it can also output bounding boxes for embedded figures/images for workflows that need lightweight layout cues. LightOnOCR-2 is released under the Apache 2.0 license, together with a small family of open-weight checkpoints (OCR-focused and bbox-capable variants, plus base checkpoints) that can be used by the community for fine-tuning, domain adaptation, and layout-oriented applications.

01/26 UPDATE: The paper is now out! It includes the full training recipe including data/normalization pipeline, RLVR, and merging details. Read it here