System One Models

System One Models is a class of AI models, defined by TypeSafe AI, built to make fast, structured decisions that software can use directly. Instead of generating text, code or reasoning explanations, the model evaluates a state and returns a typed answer together with a probability. The name evokes "System 1" — fast, intuitive thinking — as opposed to the slow, deliberative reasoning of reasoning models.

The models answer structured questions through three primitive types: Choice (selecting from predefined options), Score (a numeric rating on a scale) and Noul (a boolean true/false question). They are optimized for calibration: their probabilities reflect genuine uncertainty, measured across groups of predictions (which does not guarantee that an individual answer is correct). With calibrated confidence scores, developers can combine answers into deterministic checks in code and route cases for action, human review, or a reasoning model. The first, flagship model of this class is Jev.