BREAKING
DeepSeek-OCR-Web turns PDFs to Markdown
From upload to structured output
1
Upload PDF or image
↓
2
OCR and layout analysis
↓
3
Table and chart extract
↓
4
Markdown output
0
%
below 10x compression
0
%
at 20x compression
0
k
pages/day on one A100
OmniDocBench Edit Distance
DeepSeek-OCR 2
0.1
Gemini-3 Pro
0.115
Prior DeepSeek-OCR
0.129
Strengths and open challenges
Strengths
●
Strong table and layout reproduction
●
Fast via vLLM
●
Open source and local
Challenges
●
Degrades on long documents
●
Scanned PDF table accuracy
●
Non-deterministic output
Open-source OCR for RAG and papers
AI NEWS BLITZ
A new web tool converts PDFs and images into Markdown using the DeepSeek-OCR model.