It's about information density.

Most LLMs by default seem to write both text and code with low information density.

You can kind of get around it by prompting them to optimize for compactness, but most just let it run with a more generic prompt.