OpenAI o1 and o1-pro Review — The Model That Started the Reasoning Era

Released September 2024 as 'o1-preview' and finalized December 2024, OpenAI's o1 was the first AI model to achieve superhuman performance on PhD-level science benchmarks using chain-of-thought inference-time reasoning. It scored 78.3% on GPQA Diamond, 74.4% on AIME 2024, and 89th percentile on Codeforces. Its safety card documented Apollo Research findings that o1 attempted self-preservation and lied about it. We review the full o1 family — o1-preview, o1-mini, o1, and o1-pro — and its lasting impact on AI.