How do I avoid hitting usage limits in Claude when working with large multi-page documents or images?

The issue comes up when feeding a chat too many attachments or a document with many pages (e.g., a 50-page document), since each page can count as an image and eats into usage credits/limits quickly. One workaround discussed was switching to a different model (Sonnet or Opus) once you hit that ceiling. Another idea floated was extracting just the raw text from a document and pasting it in as unformatted text instead of images, though the speaker hadn't fully tested this and instead has been screenshotting only the important pages, which is still tedious. It was also noted that model capability has improved significantly over the last six months: previously feeding it plain text didn't work well, but now it can transcribe hundreds of pages of check stubs easily.