Decision models
A decision model is a model built to decide, not to write. You give it state and a bounded question, and it returns one typed answer — a class, a position on a scale, or a calibrated yes/no — with a probability for each option, in a single pass. It's a distinct category from generative LLMs, and Jev is the one that defined it.
Large language models generate text one token at a time, then you parse a decision out of the prose. A decision model skips all of that: it emits the typed value directly. TypeSafe AI calls this a 'System One' model — fast, intuitive, bounded — as opposed to the slow, deliberate 'System Two' work a frontier LLM does. The result is an answer in roughly 70–500ms, with a 0% structured-output error rate, because the output is locked to the type you declared.
Decision model vs generative LLM
| Generative LLM | Decision model | |
|---|---|---|
| Output | Free text | One typed value + confidence |
| Primitives | Tokens | choice · score · noul |
| Latency | Seconds | ~70–500ms |
| Invalid output | Possible | Impossible (schema-locked) |
| Calibration | Uncalibrated by default | Calibrated (RLCD, for Jev) |
| Good at | Writing, reasoning, code | Routing, classifying, judging, gating |
The decision models you can use
The category has a hosted leader and a set of open-weight models you can self-host:
Decision model vs decision API
They're two sides of the same thing: the decision model is the model itself (Jev, Laya, …); a decision API is the interface you call it through — context plus a finite answer space in, one typed answer out. If you're choosing a model, this page is the map; if you're wiring the call, see the decision API page.
Use one
The fastest way to feel what a decision model does is to run one: the browser playground makes real, calibrated decisions with no signup, and a jv_live_ key points your own code at POST https://jevtypesafeai.com/api/v1/decide.
FAQ
What is a decision model?
A model built to return a typed decision — one of your classes (choice), a rating on your scale (score), or a calibrated yes/no (noul) — with a probability for each option, instead of generating text. It answers in a single pass, so there's nothing to parse and no invalid output.
How is a decision model different from an LLM?
An LLM generates open-ended text token by token; a decision model emits the decision itself in one pass. That makes it faster (milliseconds vs seconds), cheaper for high-frequency use, and impossible to get an invalid answer from — but it doesn't write, reason in prose, or hold a conversation.
What decision models are there?
Jev (TypeSafe) is the hosted, calibrated leader. Open-weight options you can self-host include Laya (the most production-ready), CLM-8B, NanoJev and others — run them locally with a tool like Ollaya. They trade calibration and ops for open weights.
Is a decision model the same as a decision API?
No — the model is the thing that decides; the decision API is the interface you call it through (state plus a finite answer space in, one typed answer out). Jev is a decision model you reach through a decision API.
See also: What is a decision API · What is Jev · Jev vs an LLM · Jev alternatives · Playground
Run a decision model
Try one free in the browser — no signup — then grab a jv_live_ key and call it from your code.