System One Model
A model that outputs a typed decision and confidence score directly, instead of generating an answer word by word.
Definition
A System One model is trained to look at a situation and immediately output a structured decision — such as a typed answer with a probability or confidence score — instead of generating a response one word at a time the way a typical large language model does. Skipping word-by-word generation lets it respond far faster and cheaper, which suits real-time settings like games, robotics, and simulations where a fast, reliable call matters more than a conversational, explained-out answer. The name plays on the 'System 1 vs. System 2' idea from psychology, contrasting fast intuitive judgment with slower step-by-step reasoning: a reasoning model that writes out chain-of-thought is doing the slow, deliberate kind, while a System One model is built purely for the fast kind. TypeSafe AI introduced the category in September 2026 with its Jev model.