A discrepancy between first- and third-party benchmark results for OpenAI's o3 AI model is raising questions ... a challenging set of math problems. That score blew the competition away ...