我将 4 款模型的推理档位调至「高」：只解决了一个问题，却为所有消耗买了单

I Turned the Reasoning Dial to 'High' on 4 Models. It Fixed One Thing and Billed Me for Everything.

An empirical benchmark across four LLMs reveals that increasing reasoning effort to 'high' rarely improves task accuracy outside complex logic puzzles, while consistently inflating token usage and inference costs.