DeepSeek-V4-Flash-Vision-Exp

DeepSeek-V4-Flash-Vision-Exp is the experimental API model that accepts images beside text. The docs are specific: JPEG, PNG, GIF, WebP; user messages only; other models return an error if you send a picture. It can describe photos, read text from screenshots, and talk through charts. It does not generate images or video. I keep that distinction loud because people hear “vision” and expect a paintbrush.

You can send a base64 data URL, a public https link, or a file_id from the Files API. Detail can be low, high, original, or auto. Images become input tokens after a resize rule; the docs cap that conversion. Same 1M context family as the other V4 names, with Flash-like pricing on the table I opened.

I still write code against DeepSeek-V4-Flash or DeepSeek-V4-Pro when there is no picture in the prompt. Vision-exp is a specialist, and the “exp” suffix is a warning that the name can move.