OpenAI Launches o1 Model, Boosting AI Reasoning on Math, Coding and Science
Read the full article for context, quotes, and updates from the team.
OpenAI has unveiled o1, a new family of models built to improve advanced reasoning on complex problems in mathematics, coding and science. The company says the system is designed to tackle tasks that require multi-step thinking, with performance that pushes closer to expert-level problem solving.
According to OpenAI, o1 achieved strong results on several challenging benchmarks, including 83% on AIME 2024 and 74.3% on GPQA Diamond, two widely watched tests of mathematical and scientific reasoning. The company says these results reflect a major step forward in AI’s ability to handle difficult, PhD-level questions.
Unlike earlier models that often respond quickly after a single pass, o1 is trained using reinforcement learning to develop internal chain-of-thought reasoning. In practice, this means the model can spend more time “thinking” before producing an answer, allowing it to work through problems more deliberately.
OpenAI said the new approach is intended to make the model more reliable on tasks where accuracy and structured reasoning matter most. The launch underscores the growing competition among AI developers to build systems that can do more than generate fluent text, and instead reason through complex technical challenges with greater depth and consistency.
