What is a System One model?

Jev is TypeSafe AI's first System One model, launched in September 2026. A System One model returns fast, calibrated decisions instead of text.

What a System One model is

The name System One model comes from psychology's System One: fast, intuitive judgment, as opposed to System Two, the slow verbal reasoning that LLMs imitate. A System One model does the snap-decision layer of software. Jev, from TypeSafe AI, is built for exactly that job. The name itself nods to the Jevons paradox.

What a System One model is for

A System One model is called by other software, not read by people. It classifies, scores and gates in-line in 70 to 500 ms, roughly 40 to 200 times faster than a frontier LLM. Because a System One model returns typed, calibrated values, your code can act on the answer directly without parsing prose.

The three primitives of the System One model

Jev's System One model exposes three decision types you compose per call:

• choice: pick one of up to 255 options; returns the key, per-option probabilities and a confidence.
• score: an ordered 2-to-10 level scale; returns a fractional score, a distribution and a confidence.
• noul: a calibrated yes or no expressed as a probability between 0 and 1.

One call carries a model, a state (your input) and a map of questions. Several questions run in parallel in a single round trip and the answers come back keyed, so you can ask many things at once.

System One model vs System Two

An LLM imitates System Two: slow, verbal, generative. The System One model, trained with RLCD for calibration rather than fluency, does the opposite and commits to a typed answer with probabilities. Output is locked to your declared type, and confidence shows how concentrated the answer is, not that it is correct. Founded by Diogo Almeida with Erik Gafni and Sasha Sheng, TypeSafe AI shipped Jev with a $40M seed led by DCVC.

See also: What is Jev · Use cases · Get Jev

Related: Jev vs an LLM · Jev models · Jev benchmark