OpenAI, in collaboration with Microsoft (MSFT.O), has unveiled its latest innovation: the “Strawberry” series of AI models. Announced on Thursday, these models are designed to significantly improve reasoning and problem-solving abilities across complex tasks in science, coding, and mathematics.
The new “Strawberry” series includes the o1 and o1-mini models, which are now available in ChatGPT and through its API. These models represent a substantial leap forward from previous AI capabilities, reflecting OpenAI’s ongoing commitment to advancing artificial intelligence technology.
According to a blog post by OpenAI, the o1 model has achieved remarkable results, scoring 83% on the International Mathematics Olympiad qualifying exam—an impressive jump from the 13% score of its predecessor, GPT-4o. The model has also shown exceptional performance in competitive programming and has exceeded human PhD-level accuracy on science problem benchmarks.
A key innovation in the “Strawberry” series is the integration of “chain-of-thought” reasoning. This technique allows the models to break down complex problems into smaller, logical steps and refine their problem-solving strategies autonomously, similar to human reasoning processes.
Noam Brown, an OpenAI researcher focused on model reasoning, confirmed on social media that these new models are indeed the culmination of the “Strawberry” project. Brown highlighted the models’ ability to perform general reasoning tasks with unprecedented accuracy and efficiency.
The development of the “Strawberry” series was first reported by Reuters in November 2023 under the project name Q. By July 2024, Reuters revealed that the project had evolved into the “Strawberry” series, showcasing OpenAI’s continued efforts to push the boundaries of AI technology.