OpenAI

OpenAI o1 is a new family of frontier reasoning models designed to spend more time thinking before they respond, enabling stronger performance on complex tasks in science, coding, and math. In internal and third‑party style evaluations, the next update in the o1 series performs at PhD‑student levels on challenging STEM benchmarks, solves 83% of IMO qualifier problems (vs. 13% for GPT‑4o), and reaches the 89th percentile in Codeforces coding contests. The series currently includes o1‑preview and o1‑mini, the latter being a faster, cost‑efficient variant that’s particularly effective at coding and 80% cheaper than o1‑preview. Safety has been a central focus: o1‑preview shows markedly stronger jailbreak resistance than GPT‑4o, and OpenAI has introduced new safety training and formal collaborations with the U.S. and U.K. AI Safety Institutes. o1 is available in ChatGPT for Plus and Team users (Enterprise and Edu gain access next), and via the API for qualified developers. Early API access focuses on core reasoning, with some features temporarily unavailable while the models are in preview. If you’re a researcher, engineer, or builder facing multi‑step reasoning—like debugging complex codebases, deriving mathematical proofs, or annotating scientific data—o1 provides a higher ceiling for correctness and reliability than prior general‑purpose models. As the preview evolves, OpenAI plans to add browsing, file and image uploads, and automatic model selection, making o1 useful for a broader range of everyday tasks while maintaining strong safeguards.