OpenAI says newest AI models o3 and o4-mini can ‘think with images’

Opening the artificial intelligence model, who said is able to “thinking about immo”

The principal model primarily from opening is called O3, and company simultaneously released a smaller model of the or4-mini. The Rugout keeps September Debt of the first model of openay-openia, O1, which concentration on complex and deliberate complex problems on their answers.

With O3, users can load the blanks, sketches and other pictures and have the ai and discuss. Patterns can also rotate, zoom in and use other image edition instruments.

Since introducing their viral chatboot at the end of 2022, Open the update has quick to the update beyond the images in the images, voice and videos. The company is race to stay in front of the generative you, where is the heated competition from the rivals including GoogleAnthropic and Elon Musk’s they are similarly.

“For the first time, reasonable patterns can use independent using all tools of chat – Web, a chicken understanding, and the image” has written. “This helps solve complex, multi-step problems more effectively and take true steps toward acting independently.”

The company, assessed to $ 300 billion At a financing point last month he said it was o3 and o4-min their own patterns that can “think with images.” This means “you don’t see an image, that can integrate direct information directly into the reasonon,” according to the opening.

Last Month, Open Relyased a native generation function that went viral in line for their ability to produce anime images of the studio of studio.

Opening his or3 model is especially not tuned for math, coding, and understanding the addition, while o4-mini-mini-miner fastest and a cheapest cost. The two models were available to start wednesday to the chatgpt plus, pro and Team customers.

The Openai User Community has a long one of their odd pattern names or confused. CEO Samman has joined the banter this week, write in a post office On X, “How do you fix our pattern you call this summer and everyone makes a little more months to tease us (that we don’t deserve a lot) until when?”

The company also said that two models were “stress – tested under our most rigorous security program” and related to his “preparation framework is updated first week.

Open is coming under the fire of recently for changes to their security precautions and trials. Said this week that reserving the right to “change their security requirements if” another border you release a high risk without comparable abode. “”

In changing their policies this week, open it wrote it would not require safety test for some final county models. The company also avoided to release a “model card” – or a report containing information on security tests before a model is released – for their Gpt-4.1 model. I am In February, opened up agent agent tool Deep Research weeks before publish their system card.

Opening did not immediately respond to a comment request.

Wat: Open considering the building Social network