Tendem
Mini toolsNuMarkdown 8 B Thinking
By NuMind
New
Text
Gen 7
Overview
NuMarkdown-8B-Thinking is a reasoning OCR vision-language model fine-tuned from Qwen2.5-VL to convert complex document images into clean Markdown, using intermediate “thinking” tokens to infer layout and tables before generating the final text
Description
NuMarkdown-8B-Thinking is NuMind’s first reasoning OCR VLM, trained specifically for document-to-Markdown conversion for RAG and retrieval systems. It generates internal reasoning tokens to analyze layout, headers, tables and footers, then outputs structured GitHub-flavored Markdown. Built by fine-tuning Qwen2.5-VL on synthetic Doc → Reasoning → Markdown data and an RL GRPO phase, it outperforms generic VLMs and dedicated OCR models on complex, multi-column and table-heavy documents.
About NuMind
NuMind is an Artificial Intelligence company that enables software engineers, data scientists, and non-experts to effortlessly create advanced machine learning models utilizing LLMs for automated text processing.
Industry:
Software Development
Company Size:
1-10
Location:
Cambridge, Delaware, US
Website:
numind.ai
Last updated: January 26, 2026
