OpenAI has rolled out two new reasoning models, o3 and o4-mini, a major milestone in the evolution of AI's capabilities. The models are the first to have embedded agentic capabilities, enabling them to take advantage of tools like web searching, file analysis, visual reasoning, and image generation. OpenAI stated that the models are trained to figure out when and how to utilize such tools so that they are able to give more precise, comprehensive, and well-formatted responses.

This is one step toward developing a fully agentic ChatGPT capable of independently performing tasks for users. The o3 model has been outlined as the strongest reasoning model published by OpenAI to date. The model is outstanding in different fields, such as coding, math, science, and visual understanding, and especially suitable for responding to tricky, multi-step queries where answers are not explicitly known. The model is also effective in visual work such as image interpretation, charts, and diagrams.

Conversely, the o4-mini model is made to be smaller, more efficient, and quicker but with great performance in coding, mathematics, and visual reasoning. Even though it is small, it outperforms its older model, the o3-mini, particularly in data science applications and has greater usage limits. Both models are capable of interpreting and editing images, including low-resolution or blurred images, and editing visuals in real-time as part of their reasoning process.