The company finally unveiled the new system in September, outing it as OpenAI’s first “reasoning” model and renaming it “o1.” Much like the two-stage release of GPT-2, where a stripped ...