A polished search box is a weak coding-model benchmark. It can look right, return results and pass a happy-path demo while showing the wrong results to someone who types quickly. Before comparing generated screenshots, give the tools a failure they have to explain. Here is one small, reproducible example. This article does not rank current model versions: I have not run a Claude-versus-ChatGPT comparison for it. The bug hides in a reasonable-looking function let vis...