Openai says the latest AI O3 and O4-Mini models can “think”

Openai has released its latest artificial intelligence model, which, she said, is able to “think with images”, that is, it can understand and analyze the sketches and user diagrams, even if they are low in qualities.

The main new model of reasoning with Openai is called O3, and the company has simultaneously released a smaller model called O4-Mini. The release follows from the September debut of the first Openai O1 reasoning, which was focused on solving complex problems and reflection on its few steps.

Using O3 users can load boards, sketches and other images, as well as analyze II and discuss them. Models can also rotate, increase and use other image editing tools.

Since the introduction of its viral chat chatgpt at the end of 2022, Openai quickly upgraded its models to go beyond the text, voice and video. The company is racing to stay forward in the generative II, where he threatens hot competition from competitors, including GoogleAnthropic and Elon Musk equally.

“For the first time, our reasoning models can independently use all Chatgpt tools – web -styt viewing, python, understanding of images and images generation,” Openai writes. “It helps them more effectively solve complex, multistage problems and take real steps to independent.”

A company that is evaluated in 300 billion dollars In the round financing last month, they said that O3 and O4-Mini is its first AI models that can “think with images”. This means that “they don’t just see the image, they can integrate visual information directly into the reasoning chain,” Openai believes.

Last month Openai released Native Function of Image Generation It has become viral on the Internet for the ability to produce GHIBLI-Steyle anime-vios.

Openai said its O3 model is especially tuned to mathematics, coding, science and understanding of images, while O4-Mini works faster and with less costs. Both models were available since Wednesday to Chatgpt Plus, Pro and Team Clients.

The Openai user community has long joked about its amazing or confusing names of AI models. This week, CEO Sam Altman joined the banter by writing in pillar On X, “How about correcting our model before this summer, and everyone will get a few more months to make fun of us (what are we very deserve) so far?”

The company also stated that both models were “tested by stress within our strict security program today” and is related to its “readiness framework” updated earlier this week.

Openai recently fell under fire to change precautions and security processes. This week, it was said that it reserved “to change its security requirements when” another AI border developer issues a high -risk system without comparable guarantees. “

By changing his policy this week, Openai wrote that no security test for some models. The company also avoided the release of “Model Card” – or reports containing information about security tests made before the model exit – for it Model GPT-4.1. In February Openai launched AI AGENT GREAT Deep Research tool a few weeks before Publication of one’s system map.

Openai did not immediately respond to a comment request.

See: Openai with the construction of a social network