o1

OpenAI's first reasoning model, trained with reinforcement learning to perform extended chain-of-thought reasoning before answering. Uses hidden "thinking tokens" to explore multiple approaches, verify steps, and self-correct — pioneering test-time compute scaling where inference cost trades for accuracy.

o1 achieved breakthrough results on reasoning benchmarks: 83.3% on AIME 2024 (vs 13.4% for GPT-4o), ranked in the 89th percentile on Codeforces, and exceeded human PhD-level accuracy on GPQA-Diamond (78%). 200K context (full o1, December 2024). Also released as o1-mini for cost-efficient reasoning. AA Intelligence Index: 31. Proprietary.