Jinse Finance reported that the DeepSeek-AI team has released a paper titled "DeepSeek-OCR: Contexts Optical Compression," proposing a new method for compressing long text contexts using visual modalities. According to the Hugging Face page, the model has 3B parameters.