I think most thinking models can do it to be fair, I think when I tried this it was all with non-thinking models/modes. Wasn't trying to make a point that LLMs can't do it or anything, just thought it was weird.