The OpenAI o3 model is a next-generation inference model following the o1, featuring both o3 and o3-mini versions. Under certain conditions, o3 approaches capabilities of Artificial General Intelligence (AGI), scoring as high as 87.5% on the ARC-AGI benchmark test, significantly surpassing the human average. It excels in mathematics and programming tasks, achieving 96.7% in the 2024 American Mathematics Invitational Exam (AIME) and reaching a rating of 2727 on Codeforces. The o3 model can self-verify facts and use the 'private reasoning chain' for inference, enhancing answer accuracy. It is the first model trained with 'deliberative alignment' techniques to adhere to safety principles. Currently, the o3 model is not widely available, but security researchers can register for preview access to the o3-mini model. The o3 mini version will launch at the end of January, followed closely by the full version.