Image support depends on the exact DeepSeek checkpoint. OpenRouter'sSeptember 29 catalog guide lists V4.1 Flash and V4 Flash Vision Exp as accepting images, while the listed V4 Pro models accept text only. A text-only model can still reason over a description produced by another vision model, but it receives that description rather than the image itself. Developers therefore need to check the selected endpoint's input types before sending images; routing among hosts of the same model does not add vision support.
