OpenAI’s Secret “Strawberry” Project Aims to Increase AI Reasoning Abilities

OpenAI is reportedly working on a new advanced reasoning technology for its AI models, under a project, code-named, ‘’Strawberry,’’ reported Reuters last week.

Reuters, citing company’s internal documents and a knowledgeable source, reported that the project, a well-kept secret even within OpenAI, aims to empower its AI models to autonomously navigate the internet and conduct “deep research.” According to these documents, the ongoing project will involve not only generating answers for queries but also planning and executing complex tasks.

The internal document detailing the company’s usage plan for Strawberry was reviewed by the news agency in May of this year. Although Reuters could not ascertain the document’s exact date or how close the project is to public release, a source described the project as a work in progress.

Sources familiar with the matter suggest that by enhancing AI with advanced reasoning capabilities, OpenAI seeks to equip its models with skills that traditionally challenge AI systems, such as common-sense tasks and intuitive logical reasoning, abilities humans possess effortlessly.

When questioned about Project Strawberry, a OpenAI spokesperson, without directly addressing specifics of the project, said, “We aim for our AI models to perceive and comprehend the world more akin to human understanding. Ongoing research into new AI capabilities is standard in the industry, driven by the belief that these systems will evolve in reasoning abilities over time.”

For now, many aspects of project Strawberry remain undisclosed — including its development progress, and whether it’s the same project with “human-like reasoning” skills that OpenAI demonstrated at an internal meeting last week, per Bloomberg.

However, per Reuters, sources acquainted with the project revealed that Strawberry, formerly known as Q*, has demonstrated the ability to address difficult science and math questions beyond the capabilities of existing commercial-AI models.

As AI technology advances, the development of reasoning skills is seen as key for current large language models to develop human-like capabilities, leading to significant progress across various domains.