DeepSeek drops open-source model that compresses text 10x through images, defying conventions

DeepSeek-OCR, a groundbreaking AI model from China, compresses text 10x by converting it into images—redefining how language models process and remember information.