Early yesterday morning, OpenAI unveiled two new models: o3 and o4-mini. A standout feature of these models is their ability to "Think with images," meaning they can seamlessly incorporate visual data into their thought processes for reasoning. For example, upon receiving an image of the Beijing subway map, the o3 model can analyze the image to provide answers to questions like, "How do I get from Tiananmen Square to Caoqiao?" This capability showcases the model's advanced visual reasoning skills, enhancing its efficiency in tackling complex issues involving image information. On the other hand, the o4-mini is a more compact version, offering a superior cost-effectiveness ratio, making it ideal for swift interactions or smaller coding tasks. Both models support multimodal inputs and can automatically invoke the ChatGPT tool chain, marking a significant leap forward in AI technology, particularly in image processing and reasoning capabilities.