OpenAI, backed by Microsoft, unveiled its latest “Strawberry” series of AI models, designed to spend more time processing answers to handle more complex and challenging queries.
According to OpenAI, the new models—referred to as o1 and o1-mini—are capable of reasoning through intricate tasks, showing notable improvements in fields like science, coding, and mathematics compared to previous models.
The project, internally codenamed “Strawberry,” was first reported by Reuters. OpenAI announced that the o1 model would be available in ChatGPT and its API starting Thursday. Noam Brown, an OpenAI researcher specializing in model reasoning, confirmed the release of the Strawberry project on the social media platform X, expressing excitement about the models’ enhanced reasoning capabilities.
In a blog post, OpenAI highlighted the impressive performance of the o1 model, which scored 83% on the International Mathematics Olympiad qualifying exam—up from just 13% achieved by its predecessor, GPT-4o. The model also surpassed human PhD-level accuracy on scientific benchmarks and improved significantly in competitive programming tasks.
The enhanced reasoning abilities of these models stem from a technique called “chain-of-thought” reasoning, which breaks complex problems into smaller, logical steps. OpenAI has automated this method, allowing the models to approach and solve intricate problems independently, without user prompting.
“We trained these models to spend more time thinking through problems before responding, similar to how a person would,” OpenAI stated. “This process helps them refine their strategies, recognize mistakes, and ultimately improve their problem-solving abilities.”
