They do so by running the image through an OCR tool call

They can, sure...that's really just LLMs though.

ML models to recognize handwriting have existed way before LLMs could call tools, though

Identifying digits is like the "Hello World!" of ML

https://www.youtube.com/watch?v=aircAruvnKk

An OCR tool is ML. AI is generally used to mean LLM’s. You’re repeating what I already wrote

No they don't, they natively "see" images.

That's a thing I always marvel about - how LLMs are so versatile and do so much stuff so good that was out of reach just some years ago

Especially when you consider how expensive "good" OCR software is