OpenAI, supported by Microsoft, announced on Thursday the launch of its "Strawberry" series of AI models, which are designed to enhance the processing time for answering queries in order to tackle complex challenges more effectively.

According to a blog post from the AI company, these models, initially reported by Reuters, possess advanced reasoning capabilities that enable them to address more difficult problems in fields such as science, coding, and mathematics compared to earlier models.

Internally referred to as Strawberry, the newly announced models are named o1 and o1-mini. The o1 model will be accessible in ChatGPT and its API starting Thursday, as stated by the company.

Noam Brown, a researcher at OpenAI dedicated to enhancing reasoning in AI models, confirmed via a post on the social media platform X that these models align with the Strawberry project.

"I am thrilled to share the results of our efforts at OpenAI to develop AI models that can perform general reasoning effectively," Brown expressed.

In the blog post, OpenAI highlighted that the o1 model achieved an impressive 83% on the qualifying exam for the International Mathematics Olympiad, a significant improvement over the 13% score of its predecessor, GPT-4o.

Additionally, the model demonstrated enhanced performance on competitive programming tasks and surpassed human PhD-level accuracy on a benchmark for scientific problems, according to the company.

Brown explained that the models achieved these results by utilizing a method known as "chain-of-thought" reasoning, which involves deconstructing complex issues into smaller, logical components.

Researchers have observed that the performance of AI models on intricate problems tends to improve when this approach is employed as a prompting technique. OpenAI has now automated this process, allowing the models to independently break down problems without requiring user input.

"We have trained these models to take more time to contemplate problems before providing answers, similar to human thought processes. Through this training, they refine their reasoning, explore various strategies, and learn from their errors," OpenAI stated.

In November 2023, Reuters was the first to announce OpenAI's initiative on the reasoning project, initially referred to as Q*. In July, it reported that the project had since been renamed Strawberry.