Pretty wild how well GPT4 is still doing in comparison. It's significantly better than their model at creating compilable code, but is less accurate at recreating functional code. Still quite impressive.