Token

A token is a piece of a word — roughly four characters of English. Models read and write in tokens, and limits and pricing are usually measured in them.

You rarely need the exact count, but tokens explain practical limits: very long inputs cost more and can hit the context window, so trimming irrelevant text helps.

Pasting huge volumes of unrelated text "just in case". It dilutes focus, costs more, and can push the important part out of view.

Here are the two pages that matter for my question (pasted below). Ignore everything else from the report.