Vision Encoders

DeepSeek Open-Sources OCR System That Compresses LLM Contexts With Visual Tokens

By mapping text to images, the model targets long-context costs with claimed 7–20× token cuts now under community testing.