Table of Content
AI Information
AI Tutorial
AI Agent
Latest Information
Hot Information
LLMs Information
Technical Frontiers
Industry Applications
Prompt Tutorial
AI Programming
AI Video Generation
AI Audio Generation
Knowledge Base Setup
AI Development Guides
Industry Tutorials
Practical Cases
AI Resource Library
Agent Information
Agent Setup
Agent Training

0.35 seconds to OCR a full page of documents, 10% higher than Qwen2.5 VL's document conversion multimodal model!

Written by
Clara Bennett
Updated on:July-10th-2025
Recommendation

A new breakthrough in efficient document conversion, the SmolDocling model achieves extremely fast processing, and its performance exceeds that of Qwen2.5 VL.

Core content:
1. The SmolDocling model is jointly launched with IBM Research, with efficient document conversion capabilities
2. Comprehensive functional features, including OCR, layout recognition, code and formula recognition, etc.
3. Recommended reading resources, in-depth exploration of the development of AI Agents and multimodal systems

Yang Fangxian
Founder of 53AI/Most Valuable Expert of Tencent Cloud (TVP)
SmolDocling is a multimodal image-to-text model designed for efficient document conversion and is currently ranked 2nd on the huggingface hot list.
SmolDocling was jointly launched by the Docling team and IBM Research. It takes only 0.35 seconds per page on the A100 GPU on average , and its 256M parameters are more efficient than Qwen2.5 VL ( 7B )!
SmolDocling features:
DocTags efficient tagging - Introducing DocTags, an efficient and concise document representation method that is fully compatible with DoclingDocuments.
Optical Character Recognition (OCR) – Accurately extract text from images.
Layout and positioning - Preserve the document structure and bounding boxes of document elements.

Code recognition - detects and formats code blocks, including indentation.

Formula Recognition - Recognize and process mathematical expressions.

Chart recognition – extracting and interpreting chart data.
Table Recognition - Supports structured table extraction, including column and row headers.
Graphic classification - distinguishing between graphics and graphic elements.
Title Correspondence – Link titles to relevant images and graphics.
List Grouping - Properly organize and structure list elements.
Full Page Conversion - Processes the entire page for comprehensive document conversion, covering all page elements (code, formulas, tables, charts, etc.).
OCR with Bounding Boxes - OCR region recognition using bounding boxes.
General Document Processing – Trained on both scientific and non-scientific documents.
Seamless Docling integration - Import Docling and export in multiple formats.
https://hf-mirror.com/ds4sd/SmolDocling-256M-previewhttps://arxiv.org/pdf/2503.11576SmolDocling: An ultra-compact vision-language model for end-to-end multi-modal document conversion


Related PostsView More
Firebase Studio, a one-stop AI application development platform
Explore Firebase Studio, a one-stop AI development platform, which is an innovative product launched by Google. It integrates cutting-edge technologies such as Project IDX, Genkit and Gemini to provide full-process support for developers. It can not only realize natural language generation application prototypes, but also has powerful AI code generation and other functions. In-depth...
Iris Vance
Just now, Alibaba released and open-sourced Qwen3, and its benchmark indicators are quite impressive!
Just now, Alibaba released and open-sourced the AI ​​big model Qwen3, which has excellent benchmark indicators! It is highly competitive compared with many top models. Qwen3 supports a variety of tasks and application scenarios, and some model weights are open source. Want to know what open source big models are? Want to explore more of the charm of open source big models? Click to read!
Caleb Hayes
Ali Qwen3 is officially released: minimum 600 million parameters, challenging Gemini-2.5Pro
Alibaba has officially released the Qwen3 series of AI large models, which include models of various parameter sizes and are open source. Qwen3 performs well in many benchmarks and is very competitive with top models. Small models also perform well. To learn more about the open source large models and more details, click to read!
Audrey Miles