In my experience Gemini 1.5 Pro was giving me more complete code than Claude, but the Gemini-generated code would not compile (C++ code) due to several errors. I was going back and forth with Gemini and it became clear that it was in a "loop" (it would suggest a fix and then that wouldn't work and it would suggest the original code again, wash, rinse, repeat). Then when I asked Claude why that code wouldn't compile it immediately identified the problems (mainly due to libclang API changes over various versions) and the code compiled.

I think it's the exact opposite honestly. I use chatgtp because it's not a google or facebook product and this isn't a unique sentiment among my developer friends.

The linked article showed the opposite for everything except coding though..?