I compared ChatGPT 4.1 to o3 and 4o to find the most logical AI model – the result seems almost irrational