OpenAI Launches o1 Reasoning Model, Touting Major Gains on Math and Science Tasks

OpenAI has unveiled o1, a new family of models built to tackle complex reasoning tasks in mathematics, coding and science. The company says the system represents a major step forward in AI’s ability to solve multi-step problems by using reinforcement learning to “think” through challenges before producing an answer.

According to OpenAI, o1 delivered strong results on several demanding benchmarks. The model scored 83% on AIME 2024, a test of advanced mathematical reasoning, and 74.3% on GPQA Diamond, a benchmark designed to evaluate graduate-level science knowledge and problem-solving. OpenAI said these results reflect the model’s improved ability to work through difficult questions with greater accuracy and structure.

The new model family is designed for users and developers who need more than quick text generation, especially in areas where careful reasoning matters. OpenAI described o1 as a significant leap in AI systems that can handle tasks requiring multiple steps, logical consistency and deeper analysis.

Early access to o1 is now available through ChatGPT, with a broader rollout planned in the near future. The launch adds to growing competition among AI companies racing to build models that are not only fluent, but also more capable of reasoning through complex real-world problems.