Paper Notes: Deepseek-OCR

last updated 2026-05-11

vlm-ocr

Source

DeepSeek-OCR: Contexts Optical Compression

Haoran Wei, Yaofeng Sun, Yukun Li·arXiv

↓ Download PDF

TL;DR

This paper made waves when it came out, though imo those were a bit of an overreaction, just because it was from the whale. It wasn’t even a SotA OCR model when released, and the brouhaha was mostly around the core claim of the paper: you can efficiently compress prompts as document images with minimal quality loss.

joe barrow · research scientist · subscribe · rss

Enjoying this?

Get new posts and field notes by email.