Jev models and versions

The Jev models you call are the same System One engine behind one of two names: jev-latest, a moving pointer, or a pinned version like jev-1.13.0 that never shifts under you.

There is one family of Jev models from TypeSafe AI, exposed through a version string you pass on every call. Choosing between the Jev models is really choosing between tracking the newest release and pinning a fixed one, and that single choice shapes how stable your decisions are over time.

jev-latest vs pinned versions

jev-latest is a moving pointer: it always resolves to the current release, so you get improvements automatically. Pinned Jev models like jev-1.13.0 lock you to one exact version. In development, jev-latest keeps you on the newest behavior. In production, pinning is the safer of the Jev models because a calibrated model's thresholds can shift between releases, and a shift you did not plan for can quietly change which decisions clear your confidence gates.

• jev-latest — a moving pointer to the current version, best for experiments
• jev-1.13.0 — a pinned version that stays fixed for reproducible production behavior
• Pin in production so score and confidence thresholds do not drift

The System One model class

All Jev models belong to the System One class: they return typed, calibrated decisions rather than text. Across every version, the Jev models expose the same primitives — choice (pick one of up to 255 options), score (an ordered 2–10 level scale), and noul (a calibrated yes/no from 0 to 1). Every one is trained with RLCD so the confidence numbers stay meaningful. A call is a model plus state plus a map of questions, answered in parallel in a single round trip in roughly 70–500ms.

What isn't public

TypeSafe AI has not published the parameter count or model size for any of the Jev models, so treat those numbers as unknown rather than guessing. What matters in practice is the version string, not the size. Pass the same input tokens at $0.042/M (output free, about $0.0004 per decision) whichever version you pick. Send your version through the official POST https://api.typesafe.ai/v1/systemone, or via the hosted POST https://jevtypesafeai.com/api/v1/decide with a jv_live_ key. Whichever of the Jev models you name, pin it before you ship.

See also: What is Jev · How to use the API · Pricing

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