o1: the first model that 'thinks before answering'
In one sentence OpenAI ships o1-preview and o1-mini: models trained with RL on reasoning chains. On math, physics, competitive coding they beat GPT-4o by a huge margin. Paradigm shift.
OpenAI ships o1: a different kind of model. Instead of answering immediately, it "thinks" silently for seconds or minutes before writing. On your screen you see a "Thinking..." indicator for as long as needed.
Results: 83% on the USAMO math olympiad qualifier (GPT-4o was 13%). PhD-level science at "human expert" level. On competitive coding (Codeforces) it hits the 89th percentile.
A paradigm shift: not just more data and parameters in training, but more compute during inference. The "reasoning models" era begins.
Companies
OpenAI
Tools
o1-preview, o1-mini
Tags
Sources