Is there really a difference in code quality between Claude and GPT?
These days I switch between both when doing backend work. For writing short functions, honestly they seem similar, but when setting up project structure or refactoring, Claude feels like it wanders less.
But I'm not sure whether this is a real model difference or just a day when I wrote good prompts lol. What's everyone's experience?
8 answers
Agreed — Claude definitely gets less lost with this project structure.
Well, that's a bit... honestly, it's a difference in prompts, not something I'd call a difference between models. I think we should run A/B tests a few times with the same prompt before drawing conclusions.
I felt the same way. Especially when doing refactoring across multiple files, GPT often made up functions that didn't exist, while Claude tended to lose less context. Of course, as the models have been updated, that difference has narrowed a lot.