A new report published by the U.K. government says that OpenAI’s o3 model has made a breakthrough on an abstract reasoning test that many experts thought “out of reach.”