o1 models spend extra compute on internal reasoningStep-by-step logical thinking in AI models — chain-of-thought before answering hard problems. before answering — improving performance on Olympiad math, competitive programming, and multi-step analysis. They marked a shift from pure scale to inferenceRunning a trained model to produce predictions — as opposed to the training phase that learns weights.-time reasoningStep-by-step logical thinking in AI models — chain-of-thought before answering hard problems. in frontier AI products.

o1 introduced dedicated reasoningStep-by-step logical thinking in AI models — chain-of-thought before answering hard problems. models to OpenAIAn AI research company — created GPT, ChatGPT, DALL-E, and the o-series reasoning models.'s consumer and APIApplication programming interface — a defined way for programs to talk to each other or to a service. lineup.

inferenceRunning a trained model to produce predictions — as opposed to the training phase that learns weights.-time reasoningStep-by-step logical thinking in AI models — chain-of-thought before answering hard problems. changed how teams approach hard STEM and coding problems.