I wish I had something more methodical I could show. It's all subjective, but GLM-5.2 feels more human to me. Even GPT-5.6 Sol tends to be easier on the eyes for me (though the stereotype of it overengineering and no common sense are still true).

I tried using a new agent service recently and could tell immediately that it's powered by Claude due to the way it writes.