Vision Image Token Estimator

Estimate how many tokens an image costs a vision model from its dimensions, so a batch of screenshots does not surprise you on the bill.

How to use
  1. Enter the image width and height.
  2. Pick the detail level.
  3. Read the estimated token count.
Frequently asked questions
Why does an image cost tokens at all?
Because it is cut into tiles and each tile is encoded into the same context the text uses. More pixels means more tiles means more tokens.
Does resizing an image down save money?
Yes, and it is the single biggest lever. Images above the model's tile grid get downscaled anyway, so sending a 4000 pixel screenshot buys nothing.
What does the detail setting change?
Low detail uses a single coarse pass at a fixed low cost. High detail adds the tile grid on top, which is where the count grows with the dimensions.
How accurate is the estimate?
It follows the published tiling rules, so it is close. Providers change the scheme occasionally, so check against a real invoice before budgeting a large batch.
 Last updated 
Powered by maratool