OpenAI launches a new series of AI models capable of "reasoning"

The models, first reported by Reuters, are capable of reasoning through complex tasks and can solve more complex problems than previous models in science, coding and maththe AI ​​company said in a blog post.

OpenAI uses the codename Strawberry to refer internally to the project, while the models announced on Thursday are called o1 and o1-miniThe o1 will be available on ChatGPT and its API starting Thursday, the company said.

Noam Brown, a researcher at OpenAI focused on improving reasoning in the company’s prototypes, confirmed in a post on social media platform X that the models were the same as those in the Strawberry project.

“I’m excited to share with you all the fruits of our efforts at OpenAI to create AI models capable of truly general reasoning,” Brown wrote.

In its blog post, OpenAI claims that the o1 model scored 83% on the International Mathematical Olympiad qualifying exam, compared to 13% for its previous model, GPT-4o.

The model also improved its performance on competitive programming questions and surpassed the accuracy level of a human PhD on a range of scientific problems, the company said.

What can it be used for?

According to Brown, the models achieved these scores by incorporating a reasoning technique known as “chain of thought,” which involves breaking down complex problems into smaller, logical steps.

The “o1” model can be used by healthcare researchers to record data of cell sequencing, by physicists for generate complicated mathematical formulass needed for quantum optics and by developers from all fields for create and run flows “multi-step work process,” the company explained.

Researchers have observed that the performance of AI models on complex problems tends to improve when this approach is used as a stimulation technique. OpenAI has automated this ability so that models can break down problems on their own, without the user having to give them instructions.

“We’ve trained these models to spend more time thinking about problems before responding, just like a person would. Through training, they learn to refine their thinking process, try different strategies, and recognize their mistakes,” OpenAI notes.

In its blog, OpenAI stated that it is planning to provide access to o1-mini to all users. from ChatGPT Free .

With information from Reuters