Yes, and unlike the LLM they can iterate on a problem.

When I multiply, I take it in chunks.

Put the LLM into a loop, instruct it to keep track of where it is and have it solve a digit at a time.

I bet it does just fine. See my other comment as to why I think that is.