I mean, you both used the exact same prompt, how is OP 'holding it worng'?

Granted that we can't see which model was picked, but one clearly shows that thinking was enabled, the other shows it's not.

This just proves the article's point about using cheap models / free versions and complaining about the state of AI.

Well done!

It was GPT 5.2. Can't remember if I did thinking or not. Note that even after the fallacy was highlighted to the model, it seemed to have a tough time grasping what went wrong.