VisionSqueezer
Providers

Claude (Area-Based)

How Anthropic Claude bills image tokens and how VisionSqueezer minimizes area cost.

How Claude bills images

Claude (Anthropic) bills images by pixel area, not tiles:

tokens ≈ (Width × Height) / 750

There is no tile grid to snap to — every single pixel of the image, including solid-color padding, contributes directly to the token count.

Optimization strategy

Because cost scales with raw area, the winning move is to maximize useful resolution while minimizing pixel count:

  1. Strip all non-essential padding. Solid borders are pure waste under area billing.
  2. Snap-down resize only when the image exceeds what the task needs.

VisionSqueezer aggressively crops solid borders and preserves the high-information region, shaving thousands of tokens without downscaling the content that matters.

Terminal
vision-squeezer screenshot.png --model claude

Example

A standard 2400×1670 screenshot targeted at Claude:

MetricBeforeAfter
Resolution2400×16702304×1536
Tokens5,3444,718
Savings626 tokens (-11.7%)

By targeting Claude, Squeezer preserves the massive 2304×1536 resolution and solely trims solid padding — minimizing token cost via the area-based formula rather than throwing away detail.

For screenshots with solid borders, the default corner-tolerance crop is ideal. For photographic content, add --smart-crop.