DeepSeek-V4-Flash-Vision-Exp

DeepSeek-V4-Flash-Vision-Exp is the first experimental multimodal model in the DeepSeek-V4 family. It builds on the DeepSeek-V4-Flash architecture by adding visual modules (a ViT vision encoder and aligner) and continued training to unlock visual understanding.

The model operates in an image-text-to-text mode: it takes image and text input and produces text output. Its architecture is a Mixture-of-Experts (256 routed experts, 6 active per token, 1 shared), with DFlash attention, Hyper-Connections, and a DSpark speculative-decoding path. Weights are FP8-quantized and the context window reaches 1,048,576 tokens (YaRN scaling).

Compared to DeepSeek-V4-Flash-0731, it achieves substantial gains on multimodal agent tasks while maintaining comparable text-only performance. Weights are released on Hugging Face under the MIT license; the model can be self-hosted via vLLM or SGLang.