[DeepSeek Releases OCR2, Capable of Recognizing Images in the Same Logical Order as Humans] BlockBeats News, January 27, DeepSeek has released the new DeepSeek-OCR 2 model, which adopts the innovative DeepEncoder V2 method. This allows AI to dynamically rearrange various parts of an image based on its meaning, rather than mechanically scanning from left to right. This approach simulates the logical process humans follow when observing a scene. Ultimately, the model outperforms traditional vision-language models in handling images with complex layouts, achieving smarter and more causally-inferential visual understanding.