OpenAI has unveiled its latest duo of artificial intelligence models – the high-performance o3 model excelling in both programming and image analysis, and the o4-mini model designed for improved speed while maintaining high efficiency.
What sets both models apart is their ability to “think while looking at images”. They can view an image and analyze it concurrently. For instance, if presented with an upside-down image, they can zoom in, rotate it back, and automatically read any text, or when solving a maze puzzle, they can draw a path and then re-examine the image to see if the path is correct.
Another impressive feature is training the models to use specific tools like coding, web browsing, and zooming into images piece by piece. The training process involves reinforcing the models to utilize these tools through reinforcement learning techniques to achieve optimal results.
ChatGPT users will have access to the o3, o4-mini, and o4-mini-high models, while enterprise-level users will receive them within a week. API customers can utilize the models through the Chat Completions API and Responses API. Soon, the Responses API will offer additional tools like web searching, file retrieval, and code execution.
TLDR: OpenAI introduces advanced AI models o3 and o4-mini, focusing on image understanding and tool utilization, catering to various user needs and increasing functionality through reinforcement learning.
Leave a Comment