The co -founder and CEO of Sam Altman de Openai speaks, the Week of Italian Technology 2024 in Og Officine Grandi Rarazioni on September 25, 2024 in Turin, Italy.
Stefano Guidi | Getty Images News | Getty images
Operai has launched his latest artificial intelligence model, which according to him is capable or “thinking with images”, which means that he can understand and analyze a user’s sketches and diagrams, even if they are of low quality.
The new OpenAi reasoning model is called O3, and the company simultaneously launched a narrower model called O4-mini. The deployment follows the debut in September of the first OpenAI reasoning model, O1, which focused on solving complex problems and deliberating on its answers in multiple steps.
With O3, users can load slates, sketches and other images and make the AI analyze and discuss. Models can also turn, zoom and use other image editing tools.
Since he presented his Chatbot viral chatpt at the end of 2022, Openi has quickly updated his leg so that his models go far beyond the text in images, voice and videos. The company is running to stay at the forefront in the generative AI, where it faces a heated competition of rivals, including GoogleAnthrope and Elon Musk’s XAI.
“For the first time, our reasoning models can independently use all chatgpt tools: web navigation, python, image understanding and image generation,” Openii wrote. “This helps them solve complex problems and several more effective steps and take real steps to act independently.”
The company, valued at $ 300 billion in a round of funds last month, said that O3 and O4-Mini are its first AI models that can “think with images.” That means “not only see an image, but they can integrate visual information directly into the reasoning chain,” according to OpenAI.
Last month, Openai launched a native general image characteristic that became viral online due to its ability to produce Gibli Studio -style anime images.
Operai said his O3 model is specific to mathematics, coding, science and understanding images, while O4-mini operates faster and more low. Both models were available from Wednesday for Chatgpt Plus, professional and team customers.
The Openai user community has long joked about its strange or confusing names of AI models. The CEO Sam Altman joined the jokes this week, writing in a publication about X: “What if we solve our model name for this summer and all have a few more months to make fun of us (why do we deserve) until then?”
The company also said that both models had been “tested by stress in the most rigorous security program to date” and linked to its “framework prior to the penalty” updated earlier this week.
Operai has recently been criticized for changes in its precautions and security processes. He said this week that he reserved the right to “change his security requirements if ‘another border developer releases a high -risk system without comparable safeguards.” “
When changing his policies this week, Openai wrote that he would no longer require security tests for certain tight models. The company also avoided the publication of a “model card”, or a report that contains information on the security tests carried out before a model is published, for its GPT-4.1 model. In February, Operai launched Agent Tool Deep Research weeks before publishing his system card.
Operai did not immediately respond to a request for comments.
Look: OpenAI considering building a social network
