Anyone compared code quality between Claude and GPT?

Agreed, Claude is definitely stable on the backend side these days lol

Well, that's a bit... In the end, how you provide the prompt and context is 90% of it, so I wonder whether comparing models is even meaningful. Do you have any data that compares output test pass rates by running the same task 20 times each with the same spec? Without something like that, speaking from feeling alone is no different from saying it's just your imagination.