What does it cost to process an image with a vision model?