AI Reasoning Model Outperforms Doctors in Harvard Trial

OpenAI’s o1 reasoning model surpassed human physicians in a Harvard-led study published in the journal Science ^[1][2]. The AI model accurately diagnosed complex patient cases in emergency triage settings up to 82% of the time ^[1]. This performance exceeded the 70% to 79% accuracy range recorded for expert human doctors during the same trial ^[1][3].

Researchers tested the model on difficult clinical scenarios where rapid decision-making is essential for patient outcomes ^[2]. The study highlights how advanced reasoning capabilities can assist medical staff in high-pressure environments ^[2][3]. While the results are significant, the researchers emphasized that the AI serves as a clinical support tool rather than a replacement for human medical judgment ^[1][2].