# NuMarkdown-8B-Thinking

> Open-source vision-language model converting complex documents into structured Markdown using reasoning.

NuMarkdown-8B-Thinking is an open-source vision-language model fine-tuned from Qwen2.5-VL-7B designed to convert document images and PDFs into clean Markdown. It generates internal reasoning tokens to analyze document layout and table structures before producing the final text. This architecture makes it especially effective for ingesting visually complex documents into RAG pipelines.

Website: https://huggingface.co/numind/NuMarkdown-8B-Thinking

## What it is for

- Convert complex scanned PDFs and document images into structured Markdown
- Extract nested and multi-row tables cleanly for RAG ingestion
- Parse multi-section document layouts without losing hierarchy or text flow

## Features

- Generates thinking tokens to analyze layout and merged cells prior to Markdown output
- Fine-tuned on synthetic reasoning traces with GRPO reinforcement learning
- Specialized handling for irregular layouts, multi-column blocks, and complex tables
- Compatible with Hugging Face Transformers and deployable via vLLM
- Released under an open-source MIT license

## Pricing

Open source

## Details

- Type: AI model
- Category: [AI](https://xiry.app/topics/ai.md)
- Open source: yes
- Platforms: Self-hosted, Code library, API
- For: AI engineers, RAG developers, and researchers processing complex documents
- Alternative to: OCRFlux-3B, GPT-4o, Gemini 2.5 Flash
- Tags: ocr, vlm, markdown, qwen, rag, reasoning

## Why it was saved

- [@askgpts, 25 January 2026](https://www.instagram.com/p/DT768S5D0CE/): Accurately extracts text, tables, and document structures from complex PDFs, scans, and screenshots directly into clean Markdown for RAG pipelines and automation.

## Compare

- [Alternatives to GPT-4o](https://xiry.app/alternatives/gpt-4o.md)

## Related

- [stepfun-ai/Step3-VL-10B](https://xiry.app/resources/stepfun-ai-step3-vl-10b): Open 10B multimodal model with parallel reasoning competitive with frontier systems (AI model, Open source)
- [WeVisDoc](https://xiry.app/resources/wevisdoc): Open-source vision-language model converting complex document images to structured Markdown (AI model, Open source)
- [Unlimited-OCR](https://xiry.app/resources/unlimited-ocr): Open-source vision-language model for parsing multi-page documents in one pass (AI model, Open source)
- [LLMs.txt generator](https://xiry.app/resources/llms-txt-generator): Generates llms.txt and llms-full.txt markdown files from any website URL. (Web app, Freemium)
- [Reducto](https://xiry.app/resources/reducto): Agentic document parsing and extraction platform for AI pipelines. (Web app, Freemium)
- [Website Content Crawler](https://xiry.app/resources/website-content-crawler): Apify web crawler that extracts clean text and Markdown for LLMs (Web app, Freemium)

Updated 3 October 2026. Page: https://xiry.app/resources/numarkdown-8b-thinking

Make your own library: Xiry takes your saved reels, tweets, Reddit bookmarks and Watch Later videos, makes them searchable and pulls the useful stuff out, like this library. It's invite-only for now; join the waitlist at https://xiry.app/#waitlist.