OpenAI uses GPT-5 as an example, indicating that this model performs poorly on traditional evaluation rankings because it is "not confident enough," while its hallucination rate is significantly lower ...