OpenAI announced on Thursday the launch of its new “Strawberry” series of AI models, which are designed to spend more time processing answers to tackle complex problems.
The models, previously referred to internally by the code name Strawberry, include the newly announced o1 and o1-mini. The o1 model will be available in ChatGPT and its API starting today, according to OpenAI’s blog post.
Noam Brown, an OpenAI researcher focused on enhancing reasoning capabilities in AI, confirmed on the social media platform X that these models are indeed part of the Strawberry project. “I’m excited to share with you all the fruit of our effort at OpenAI to create AI models capable of truly general reasoning,” Brown wrote.
According to OpenAI, the o1 model scored 83% on the qualifying exam for the International Mathematics Olympiad, a significant improvement over the 13% achieved by its predecessor, GPT-4o. The model also showed enhanced performance on competitive programming questions and surpassed human PhD-level accuracy on a science problems benchmark.
Brown explained that these improvements are due to the incorporation of “chain-of-thought” reasoning, a technique that breaks down complex problems into smaller, logical steps. Researchers have observed that AI model performance improves on complex problems when this approach is used as a prompting technique. OpenAI has automated this capability, enabling the models to decompose problems independently without user prompts.
“We trained these models to spend more time thinking through problems before they respond, much like a person would. Through training, they learn to refine their thinking process, try different strategies, and recognize their mistakes,” OpenAI stated.
Reuters first reported OpenAI’s work on the reasoning project, initially called Q*, in November 2023, and later identified it as Strawberry in July.
