TechFlow News: On January 27, DeepSeek released its new DeepSeek-OCR 2 model, which adopts the innovative DeepEncoder V2 method—enabling AI to dynamically reorder image components based on their semantic meaning, rather than mechanically scanning left-to-right. This approach mimics the logical process humans follow when observing scenes. As a result, the model outperforms traditional vision-language models in processing layout-complex images (e.g., documents or charts), achieving more intelligent and causally-reasoning visual understanding. (Jinshi)
Add to Favorites
Share to Social Media
