At 1 a.m. today, OpenAI conducted a technical live broadcast and released its strongest and smartest models, o4-mini and the full-blooded version of o3. o4-mini and o3 are two multimodal models that can process text, images, and audio at the same time, and can automatically call tools such as network search, image generation, code parsing, and deep thinking mode (thinking images in the thinking chain) as agent intelligent bodies. According to the test data released by OpenAI, o4-mini reached 93.4% and 92.7% in AIME 2024 and 2025 respectively, which is stronger than the full-blooded version of o3, becoming the model with the highest accuracy rate at present; it reached 2700 points in the Codeforces test, becoming one of the top 200 strongest programmers in the world.