DeepSeek-OCR is out 🔥 huggingface.co/deepseek-ai/D… ✨High-accuracy OCR - MIT license ✨Fast GPU inference (FlashAttention 2, BF16) ✨Docs > Markdown ✨Works with transformers

Oct 20, 2025 · 8:54 AM UTC

24
162
12
1,211
I need list of companies who will be using this.
Replying to @AdinaYakup
Its the best lol 😂
2
Replying to @AdinaYakup
Using the prompt to check if I get an output in JSON instead of plan text🤞
1
Replying to @AdinaYakup
Super Exciting!!
1
Replying to @AdinaYakup
@grok in terms accuracy, specifically numbers in scanned financial documents, how does this compare with Tesseract and other open source solutions?
Replying to @AdinaYakup
Wooooowww
Replying to @AdinaYakup
comparison to qwen3 vl?
Replying to @AdinaYakup
DeepSeek-OCR: finally, a way to turn your 800-page legacy PDFs into Markdown that three new SaaS apps can compete to rebrand. Next feature request—auto-detecting whether the scanned docs are actually proprietary or just another stack of startup NDAs masquerading as innovation.
Replying to @AdinaYakup
@grok any chance we will see this in Ollama or vLLM soon?
Replying to @AdinaYakup
@grok can we prompt the OCR model to extract informatiom from a document
Replying to @AdinaYakup
Open source repo?
Replying to @AdinaYakup
@grok can i run it with 8gb vram ?
Replying to @AdinaYakup
Is someone really tired with STEM document.
Replying to @AdinaYakup
how to use this via API?
Replying to @AdinaYakup
Lit💥🚀
Replying to @AdinaYakup
@grok explain, what is that?
@grok is this surpassing MistralOCR ?
Replying to @AdinaYakup
@grok what can I use this for?