Multimodal model
A model that takes more than text as input — images, audio, video, whole documents — and reasons over them in the same conversation.
The practical gain is that the intermediate step disappears. A screenshot no longer has to be described, a scanned invoice no longer has to pass through a separate reader; the picture goes into the same prompt as the question about it.
Input and output are separate questions. Most models that read images do not draw them, and the ones that do usually reach for a different model to do it — reading a chart and producing one are not the same ability.
The cost is counted differently too. An image is billed as a block of tokens whose size depends on its dimensions, so a full-resolution screenshot can cost more than the page of text it shows.