The o-series is OpenAI’s line of reasoning models, introduced with OpenAI o1 (and an o1-preview release) in 2024. OpenAI’s “Learning to reason with LLMs” announcement describes models trained with reinforcement learning to produce a long internal chain of thought, spending more compute “thinking” before they answer, which improves performance on hard math, coding, and science problems.
The o1 announcement establishes the core idea that distinguishes this line from the standard GPT models: the model reasons step by step internally rather than answering immediately. Successor reasoning models followed in this line.
As of June 2026, OpenAI’s general models page does not list standalone o-series entries as the primary lineup; reasoning capability appears to be folded into the current GPT-5 generation. Treat the o-series as the origin of OpenAI’s reasoning approach rather than a separately marketed current product, and consult OpenAI’s live pages for the present reasoning offerings.
Why business readers should care: the o-series marked the industry shift to “reasoning” models that trade speed and cost for accuracy on complex tasks, a tradeoff that now shapes how AI is priced and which jobs it can reliably do.