DeepSeek's first visual model open source

On August 31st, DeepSeek opened the DeepSeek-V4-Flash-Vision-Exp weight. This is an experimental visual model based on V4-Flash, which adds a graphic understanding capability that allows you to view a screenshot, read a chart, and then combines tools to perform the Agent task. For the pure Agent task, it remains comparable to V4-Flash. Previously, the model could only be used through API, and now the developers can download weights for their own deployment。