o1

O1 is OpenAI's first-in-series reasoning model, leveraging a 'chain-of-thought' mechanism to solve complex, multi-step problems in STEM and coding.

O1 is a generative pre-trained transformer model from OpenAI, specifically engineered for advanced reasoning tasks: it effectively ‘thinks before it answers.’ This is achieved via a proprietary reinforcement learning approach that generates a long, internal chain of thought before providing the final output. The technology excels in technical domains like math and coding, demonstrably outperforming predecessors; for example, it scored 83% on the American Invitational Mathematics Examination (AIME) compared to GPT-4o’s 12%. The family includes the high-performance o1-preview and the cost-efficient o1-mini, providing targeted solutions for developers requiring deep, logical problem-solving capabilities.

https://platform.openai.com/docs/models/o1