OpenAI Announces "o1," a Reasoning Model That "Thinks Before It Answers"

OpenAI announced a new line of models, "o1-preview" and "o1-mini," that spend time thinking carefully before responding. They far outpaced GPT-4o on problems at the level of the International Mathematical Olympiad.