On Friday OpenAI released its next-generation inference model, o3, the latest version of the o1 "inference" model released earlier this year. Just like o1, o3 is a family of models, and o3-mini is a smaller lite model that is specifically fine-tuned for specific tasks. OpenAI claims that o3 can come close to achieving AGI, at least under certain conditions. According to OpenAI President Greg Brockman, our latest inference model, o3, is a breakthrough, with step function improvements on our most difficult benchmarks. We now begin safety testing and red team walkthroughs.