OpenAI’s new o3 and o4-mini models are all about ‘thinking with images’

Two days later To declare GPT-4.1Openai is not one, not free from two new models. The company has announced today O3 and O4-mini existence of the public. Previous, Openai says that O3, coding, math and science tasks still show the most advanced model of thinking, showing “strong performance”. As for the O4-mini, Openai is sent to the same area as a lower price alternative to the “effective results” in the same areas.

More carefully, both models offer novel opportunities not found in Openai’s past systems. For the first time, the company may use and combine all of the company’s justification models, including chapppt Web trip and generation of imagery. The company allows this ability to solve O3 and O4-mini, difficult, multi-step problems more effectively and “take real steps to act independently.”

At the same time, he could not see the images of O3 and O4-mini, but also to interpret the visual processing abilities in a significantly extending and “think”. For example, you can download pictures of white boards, diagrams or sketches – even quality qualities and new models will understand them. They can adjust the images as part of what they cause.

“The most modern substantiation has become a stronger performance between the integrated power, academic criteria and real-world tasks, and it determines a new standard in intelligence and useful,” Openai.

Also, Openai releases a new coding agent (To the clod code) Codex cli called. Developers are designed to give an interface to be able to use Openai models with local code. Works outside the box, works with O3 and O4-mini, works as support for GPT-4.1.

Today’s announcement Openai CEO Director General Sam Altman said the company changed the course on the road map Detailed in February. At that time, Altman said he would not leave Openai’s first O3 last year last yearas a standalone product. However, in early April, he declared the “change of plans”, O3 and O4-Mini’nin release and Openai pointed out to move forward.

“There is a bunch of reason for this, but the most interesting thing is that we will be able to make GPT-5 better than before,” he said wrote in x. “We also want to make sure that we would have to combine everything smoothly. And we want to make sure that we have enough opportunities to support our unprecedented demand.”

This means that in February, Rashman, probably, until the release of the GPT-5, the next “a few months,” he said.

Meanwhile, Chatgpt Plus, Pro and Team users can start using O3 and O4 starting from today. In the next few weeks, Openai will bring a stronger version of the flagship model, a stronger version of the flagship model, and will provide a stronger version and present Pro subscribers. Currently, those users can continue to use O1 PRO.