DeepSeek has introduced an experimental version of V4 Flash that can combine written instructions with images. The model is aimed at developers and can be used to describe photographs, interpret charts, and obtain information from screenshots.
What can it analyze?
The new model is called DeepSeek-V4-Flash-Vision-Exp. Its main difference from the standard version of V4 Flash is the ability to analyze visual content alongside a written question or instruction.
For example, an application could send a screenshot and request an explanation of the visible elements. It could also provide a chart for a summary or a photograph for a general description.
DeepSeek says the model maintains text capabilities similar to V4 Flash and improves performance in agent tasks that require visual understanding. These comparisons come from the company’s own evaluations and may vary depending on the use case.
Three ways to send images
Developers can include an image directly in a request, provide a public address for DeepSeek to download it, or upload the file in advance and reuse it through its identifier.
These options support local images, files available online, and content previously stored on the platform.
Supported formats
The model supports JPEG, PNG, GIF, and WebP images. DeepSeek checks the actual contents of a file to recognize its format instead of relying only on its filename extension.
The documentation recognizes GIF files, but it does not confirm that the model analyzes an entire animation as video. GIF compatibility should therefore not be confused with full audiovisual sequence analysis.
It understands images but does not generate them
DeepSeek-V4-Flash-Vision-Exp is designed to understand images. The official documentation does not present it as an image generator or photo-editing tool.
This also does not mean it will perfectly recognize every piece of text, chart, or object in an image. Results may vary depending on file quality, resolution, and the clarity of the instruction.
Experimental availability
The model is available through the DeepSeek API under the identifier deepseek-v4-flash-vision-exp.
Its name ends in Exp because it remains experimental. Its capabilities, limits, and access methods may change before DeepSeek releases a stable version. The official sources do not confirm that the feature is automatically enabled for every web or mobile app account.
