DeepSeek launches image recognition mode for multimodal understanding

DeepSeek's multimodal researcher Xiaokang Chen announced today that the image recognition mode is now live on both web and app. This new mode sits alongside the existing 'Fast' and 'Expert' modes, allowing users to upload images for the model to understand and analyze, going far beyond simple text extraction.

On the app, the image recognition mode still shows an 'under testing' message, while the web version is fully accessible. DeepSeek also shared technical details of its multimodal model in April, introducing a core framework called 'Thinking with Visual Primitives' to power this feature.