Models
Multimodal
A multimodal model accepts more than one kind of input — commonly text plus images, sometimes audio or video.
Images are converted into tokens like everything else, which is why they appear on your bill in the same unit. A single high-resolution image can cost more than a page of text.
Providers usually cap the number of images per request, and that limit is often reached before the token limit.
In practice
Images are billed as tokens, and the count depends on resolution — a high-detail image can cost more than a page of text. The expensive mistake is sending full-resolution screenshots when a downscaled version answers the question equally well. Count the image tokens before assuming vision is a rounding error on the bill.
Common questions
How are images billed?
As tokens, with the count depending on resolution and detail setting. A high-detail image can cost more than a page of text, so downscaling before sending is a real saving when the question does not need full resolution.
Can I count image tokens before sending?
Approximately, from the documented resolution formula for each provider. It is arithmetic rather than tokenization, so the figure is reliable — unlike text estimates for models without a public tokenizer.