OpenAI’s o3 model solved nearly 72% of coding problems, a steep jump from an overall high score of 4.4% in 2023, an analysis from the Stanford Institute for Human-Centered AI found.