PANews reported on January 27th that DeepSeek has released a new DeepSeek-OCR 2 model, employing the innovative DeepEncoder V2 method. This allows AI to dynamically rearrange the various parts of an image based on its meaning, rather than simply scanning mechanically from left to right. This approach simulates the logical flow that humans follow when viewing a scene. Ultimately, this model outperforms traditional visual-language models when processing images with complex layouts (such as documents or charts), achieving a more intelligent and causal reasoning-driven visual understanding.