[DeepSeek's First Visual Model Open Source] Beating AI News: DeepSeek has released the weights for DeepSeek-V4-Flash-Vision-Exp. This is an experimental visual model based on V4-Flash, featuring enhanced image comprehension capabilities, such as analyzing screenshots and reading charts, and combining tools to complete Agent tasks. For pure text-based Agent tasks, it still maintains performance comparable to V4-Flash. Previously, the model could only be accessed via API, but now developers can download the weights for self-deployment. [Original Link]