Sep 2024
OpenAI o1 — Reasoning Models That Think Before Answering
OpenAI released o1, a reasoning model that spends additional compute on internal chain-of-thought 'thinking' steps before producing its final answer, trading latency for accuracy on complex tasks. o1 scored 83% on International Olympiad in Informatics competition problems and performed at PhD level on the GPQA science benchmark covering biology, chemistry, and physics. It introduced inference-time compute scaling as a distinct axis from model size for improving AI capability.
Why it mattersChain-of-thought reasoning at inference time — rather than only at training time — made o1-class models substantially better at multi-step math, code, and logic without retraining.
Try itCompare o1 and GPT-4o on your hardest 50 production prompts: reasoning models often outperform on complex tasks but cost 5–20x more, so the routing decision matters.