DeepSeek is releasing OCR2, which can recognize images in the same logical order as humans.

BlockBeats News, January 27th, DeepSeek released the brand-new DeepSeek-OCR 2 model, adopting the innovative DeepEncoder V2 approach, enabling AI to dynamically rearrange various parts of an image based on their semantic meaning, instead of mechanically scanning from left to right. This approach simulates the logical flow that humans follow when viewing a scene. Ultimately, the model outperforms traditional vision-language models when dealing with images with complex layouts, achieving a more intelligent visual understanding with enhanced causal reasoning capabilities.