AI BriefingNuExtract3: 4B-scale VLM for Document Extraction Released
Key point
Numind unveiled NuExtract3, a 4B VLM specialized in Markdown conversion of documents and structured data extraction.
Details
NuExtract3, released by Numind under the Apache-2.0 license, is a 4B-scale open-weight vision language model (VLM) based on Qwen3.5-4B. It is designed to make information extraction tasks from complex documents practical.
Key features include:
- Converting document images to Markdown
- Structured data extraction using JSON templates
- Handling tables, forms, and pages with complex layouts
- OCR functionality supporting both text and visual inputs
Trained for 3 days on an 8xH100 node, it excels at long-context processing and can run on just 4GB VRAM. It is provided in various formats and quantized versions including Safetensors, GGUF, and MLX, making it easy to use with vLLM, llama.cpp, and more.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Apple Releases LensVLM-9B for Processing Compressed Text Images