Thinking can hurt one-shot websites and games.
We looked at the two highest reasoning effort levels of @AIatMeta's Muse Spark 1.3, Xhigh and Max. In a sample of about 800 website and game prompts that both settings answered, Max thought about a third longer than Xhigh on the websites and nearly twice as long on the games.
But when we opened every page in a browser, the way users see them, 21% of Max's pages threw an error, compared with 15% of Xhigh's.