48,000 characters in 2,700 tokens: lets discuss how LLMs read text as images

A recent experiment with the Claude Code model showed that converting text to images before passing it to the model can significantly reduce costs. This is because image tokens are priced by pixel, not character, resulting in a 89% reduction in tokens for a 48,000 character block. This technique can be useful for anyone running agents or long-context pipelines. To replicate this, text needs to be converted to images and then passed to the model.