Anyone else noticing a difference in code review quality between Claude and GPT?
Lately my company has been using LLMs for code reviews, and Claude's review comments feel sharper, while GPT's explanations feel more friendly. What criteria do you all use to choose between them?
8 answers
I just run both lol. Claude misses less, and GPT explains things better, so using them together works pretty well.
Agreed — especially when it comes to pointing out security issues, Claude feels like it catches them better.
Well, isn't that a bit subjective? It's unclear whether "sharp" means it actually finds more bugs or that its comments have a harsh tone. Comparing just three PRs could be a coincidence, so I think it would be better for the team to measure something like the number of issues caught before merge over about a month and then decide based on that.