OpenAI upgrades its smartest AI model with improved reasoning skills

OpenAI Today it announced an enhanced version of its most powerful capabilities artificial intelligence Model so far – model that takes longer to deliberate on questions – after only one day Google It announced its first model of this type.

The new OpenAI model, called o3, replaces o1, which the company created It was introduced in September. Like o1, the new model spends time thinking about a problem in order to provide better answers to questions that require step-by-step logical thinking. (OpenAI chose to skip the “o2” moniker because it is already the name of a UK mobile carrier.)

“We see this as the beginning of the next phase of AI,” Sam Altman, CEO of OpenAI, said in a live stream on Friday. “You can use these models to do increasingly complex tasks that require a lot of thinking.”

OpenAI says the o3 model scores significantly higher on several metrics than its predecessor, including those that measure complex skills related to programming and advanced proficiency in mathematics and science. It is three times better than o1 at answering the questions it asks ARC-AGIa benchmark designed to test the ability of AI models to reason about very difficult mathematical and logical problems that they encounter for the first time.

Google follows a similar line of research. Noam Shazier, a Google researcher, yesterday It was revealed in a post on X The company has developed its own thinking model, called Gemini 2.0 Flash Thinking. Google CEO Sundar Pichai called it “our most thoughtful model yet” in his book Special function.

The two dueling models show that the competition between OpenAI and Google is fiercer than ever. It is crucial that OpenAI proves it can continue to make progress as it seeks to attract more investment and build a profitable business. At the same time, Google is desperate to show that it remains at the forefront of AI research.

The new models also show how AI companies are increasingly looking beyond simply scaling up AI models in order to extract greater intelligence from them.

OpenAI says there are two versions of the new model, o3 and o3-mini. The company has not yet made the models available to the public, but says it will invite strangers to come forward and test them. Today, OpenAI also revealed more details about the techniques used to align o1. This includes having a typical reason about the nature of the application being made to question whether it might violate its guardrails.