Which Claude model should a UX researcher actually use?

I used to run everything on the top model but decided to test how big of the difference in the output for the user research work?