Comment by finnborge

If you haven't already seen the DeepSeek OCR paper [1], images can be profoundly more token-efficient encodings of information than even CSVs!

[1]: https://github.com/deepseek-ai/DeepSeek-OCR/blob/main/DeepSe...