The DeepSeek multimodal model has arrived, the technical report has been published.

DeepSeek released a multimodal model and technical report on Github. DeepSeek proposes an innovative reasoning framework based on visual primitives, integrating spatial tagging into thinking. Its model is based on a highly optimized architecture, with high efficiency in visual tagging, comparable to cutting-edge models in benchmark tests, pointing the way for the development of multimodal intelligence.