The “thinking” models are really verbose output models that summarise the thinking at the end. These tend to outperform non-thinking models, but at a higher cost.

Anthropic lets you see some/all of the thinking so you can see how the model arrived at the answer.