DeepSeek OCR Just Blew Up Document AI: How a “Vision Hack” Compresses Text 20×
1. What is DeepSeek OCR? DeepSeek‑OCR is an open‑source vision‑language model (VLM) designed for high‑throughput OCR and for solving the long‑context problem in large language models through vision‑text (optical) compression. arXiv At first glance it looks like “just” an OCR engine: feed