By OpenAI 's own testing, its newest reasoning models, o3 and o4 -mini, hallucinate significantly higher than o1.

OpenAI CEO Sam Altman has shared that users’ polite habits — specifically, saying “please” and “thank you” while interacting ...