PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model

TL;DR

The paper tackles efficient, multilingual document parsing by decomposing the task into fast layout analysis and a compact vision-language recognizer. It introduces PaddleOCR-VL-0.9B, a NaViT-based visual encoder paired with ERNIE-4.5-0.3B, and a dedicated PP-DocLayoutV2 for layout with a reading-order pointer network, enabling accurate recognition of text, tables, formulas, and charts across 109 languages. A rigorous data pipeline—combining 30M training samples, automatic annotation, and hard-case mining—drives robust cross-domain performance, and extensive benchmarks (OmniDocBench v1.0/v1.5, olmOCR-Bench, in-house data) verify SOTA results with fast inference. The work demonstrates strong potential for real-world deployment, especially in multilingual settings and RAG-enabled systems, by delivering high accuracy with efficient computation. Overall, PaddleOCR-VL advances practical multimodal document parsing through architectural efficiency, large-scale data curation, and comprehensive evaluation across diverse document types and languages.

Abstract

In this report, we propose PaddleOCR-VL, a SOTA and resource-efficient model tailored for document parsing. Its core component is PaddleOCR-VL-0.9B, a compact yet powerful vision-language model (VLM) that integrates a NaViT-style dynamic resolution visual encoder with the ERNIE-4.5-0.3B language model to enable accurate element recognition. This innovative model efficiently supports 109 languages and excels in recognizing complex elements (e.g., text, tables, formulas, and charts), while maintaining minimal resource consumption. Through comprehensive evaluations on widely used public benchmarks and in-house benchmarks, PaddleOCR-VL achieves SOTA performance in both page-level document parsing and element-level recognition. It significantly outperforms existing solutions, exhibits strong competitiveness against top-tier VLMs, and delivers fast inference speeds. These strengths make it highly suitable for practical deployment in real-world scenarios. Code is available at https://github.com/PaddlePaddle/PaddleOCR .

Figures (47)

• Figure 1: Performance of PaddleOCR-VL on OmniDocBench v1.0 and v1.5.
• Figure 2: The overview of PaddleOCR-VL.
• Figure 3: Architecture of layout analysis model.
• Figure 4: Architecture of PaddleOCR-VL-0.9B.
• Figure 5: The construction process of training data for PaddleOCR-VL-0.9B.