OpenAI has officially launched two new AI reasoning models, o3 and o4-mini. The o3 is positioned as the "most powerful reasoning model," while the o4-mini is known for its "compact size and high efficiency," striking a balance between cost and performance.

The new models introduce image reasoning capabilities for the first time, allowing "direct integration of images into the thought process." This means users can trigger model reasoning by showing sketches or whiteboard content. Additionally, the model can zoom in, rotate, and manipulate images during analysis. This feature will expand the application of AI in visual problem-solving scenarios.

In addition to image processing, OpenAI has also deeply integrated the reasoning models with ChatGPT tools. The new models will fully support extended functions such as web browsing and image generation. The first batch of tools has been made available to ChatGPT Plus, Pro, and Team users, covering three versions: o3, o4-mini, and o4-mini-high. The o3-pro version is expected to be adapted within "a few weeks," while older models like o1 and o3-mini will gradually phase out of related service tiers.

This release follows OpenAI's announcement of the GPT-4.1 flagship model on Monday, which is considered an iteration of GPT-4o. By continuously optimizing the technical architecture, OpenAI is accelerating the development of an AI capability matrix that spans multimodal and multitool functionalities.