> My guess is that Claude is trained to bias towards making minimal edits to solve problems.

I don't have the same feeling. I find that claude tends to produce wayyyyy too much code to solve a problem, compared to other LLMs.