Tinker1280-pixel screenshots: 1,334 tokens a look.
Chapter 6 of 7
Every look costs
Finance asks what one re-order costs. Every look the agent takes is an image the model has to read.
This chapter: How big should a screenshot be?
About this chapter
You will learn
- How a screenshot becomes tokens
- Why smaller screenshots cost less
- Why too small makes the agent miss
Words you’ll meet
- Visual token
- One small square patch of an image, read by the model as one token.
- Screenshot size
- The width and height, in pixels, of the picture the host sends.
Step 1 of 3
Every screenshot is an image the model reads, cut into 28-pixel squares. Each square is one .
Predict first.