DeepSeek Launches Groundbreaking Vision Model with Open Weights!

DeepSeek has released V4-Flash-Vision-Exp, a multimodal model with 305 billion parameters, enabling image processing for the first time. Available under MIT license, developers can self-host and fine-tune without Chinese regulations. This release follows the recent Z.ai model debut. The new architecture maintains strong text capabilities while outperforming its predecessor in multiple benchmarks.