OpenAI Unveils o1 Reasoning Model, Setting New Benchmarks in Math and Science

OpenAI has introduced o1, a new family of models built to tackle complex reasoning tasks in math, coding, and science. Unlike earlier systems that prioritize speed, o1 uses a “test-time compute” approach, spending more inference time on difficult problems to improve accuracy and step-by-step reasoning.

According to OpenAI, the model has delivered major benchmark gains, including scores of 83% on AIME and 78% on GPQA, outperforming previous leaders such as GPT-4o on several advanced evaluation sets. The company says the model is designed to better handle problems that require multi-stage logic, careful analysis, and deeper domain understanding.

The release marks a notable shift in how large language models are optimized, with more emphasis on deliberate reasoning rather than instant responses. OpenAI says early access to o1 is now available through ChatGPT, while broader API availability is expected soon.

The launch comes amid intensifying competition in AI, as developers race to build systems that can move beyond fluent text generation and into more reliable problem-solving. For researchers, engineers, and enterprise users, o1 signals a push toward models that can support higher-stakes work where accuracy matters as much as speed.