OpenAI, the developer of ChatGPT, is developing a secret AI project codenamed “Strawberry” that could significantly enhance AI inference capabilities and lead to breakthroughs in autonomous research and problem-solving.
Details are being kept closely guarded, but Reuters has learned, through sources and internal documents, that Strawberry takes a novel approach to training and processing AI models, allowing them to perform tasks that existing systems couldn’t.
Unlike current models that focus primarily on generating text-based responses, Strawberry aims to equip AI with the ability to “plan ahead” and navigate the internet autonomously, and to carry out what OpenAI calls “deep research,” which represents a major leap forward as it requires a deeper understanding of context, logic, and multi-step problem-solving.
The pursuit of human-level AI reasoning is a central focus across the industry, with companies like Google, Meta, and Microsoft exploring various approaches. Experts believe that achieving this breakthrough will unlock the potential for AI to drive scientific discovery, develop complex software, and tackle challenges that currently require human intuition and planning.
OpenAI has not publicly confirmed details about Strawberry, but a company spokesperson told Reuters: “We want our AI models to see and understand the world in the same way humans do. Continuous research into new AI capabilities is common practice in the industry, and there is a shared belief that these systems will improve their reasoning capabilities over time.”
Strawberry appears to be an evolution of an earlier OpenAI project known as Q*, which attracted attention within the company for its advanced reasoning capabilities. Sources who have seen a demonstration of Q* say it can solve complex math and science problems that are beyond the capabilities of current commercially available AI.
While the exact mechanisms remain secret, sources say Strawberry involves a specialized form of “post-training,” the process of refining an AI model after it has been trained on a massive dataset. This post-training stage, which likely involves techniques such as “fine-tuning” and self-generated training data, is crucial for honing the AI’s reasoning capabilities.
