Loading decision models…
THE CATEGORY
An AI decision model reads an input and returns an answer from a set you define (an option, a score, or the probability that something is true), with a calibrated probability for every allowed answer, in one pass and without generating text.
They are the fast half next to a language model: the component that decides, while the LLM writes. Since TypeSafe AI released Jev in September 2026 they are also called System One models, and 26 of them, open and hosted, are on this registry.
Also called: AI decision models · typed decision models · System 1 models · Jev-style models · Jev alternatives
| Compared with | What that does | What a decision model does |
|---|---|---|
| A large language model (LLM) | Generates text token by token. The answer has to be parsed, can come back off-schema, and takes longer the more it writes. A probability, if you ask for one, is more text. | Selects from the answers you supplied in one forward pass, usually in milliseconds. It cannot answer outside the set, and every answer comes with its probability. |
| A classifier | Also returns a label with a probability, but its labels are fixed when it is trained. A new question means new training data and a new model. | Takes the question and its allowed answers with each request. The same model routes a ticket on one call and scores a lead on the next, with no retraining. |
| JSON mode or structured outputs | Constrains an LLM’s output to a schema, so the JSON parses. It is still generation: just as slow, and the value inside has no calibrated confidence. | Returns a distribution over the allowed answers, calibrated so that a threshold means what it says: answer at 0.9, hand the rest to a person or a bigger model. |
The full argument, with an example of each: System One models vs LLMs.
The pattern that makes them safe to ship is confidence gating: act on the answer above a threshold, and send everything else to a person or a larger model.
Every model in one table, with sizes, licences, prices and reported benchmarks.
An AI decision model reads an input and returns an answer from a set you define (an option, a score, or the probability that something is true), with a calibrated probability for every allowed answer, in one pass and without generating text.
Yes, a particular kind. A System One model is a decision model: it reads a state and answers typed questions — choose an option, score on a scale, or give the probability that a condition holds — in one forward pass, with calibrated confidence and no generated text. The name comes from TypeSafe AI, which released the first one, Jev, in September 2026. People also call them AI decision models, typed decision models, System 1 models, Jev-style models and Jev alternatives.
Many are built from one: a language model’s encoder or decoder with a decision head in place of text generation. But it is used differently. It does not write; it scores the answers you supply and returns the most likely one with its probability, in a single forward pass.
A classifier’s labels are fixed at training time. A decision model takes the question and its allowed answers with every request, so one model can answer questions it was never trained on, and its probabilities are calibrated for gating.
On System One Models (systemonemodels.tech), a registry built only for them: 26 from 25 makers, 25 with open weights and 4 served as a hosted API (bespoke-nimble-9b, gliner2-5-decide, tev1 and jev). Every one has a page with what it decides, its licence, size, price and reported benchmarks, and a playground.
Yes, most of them: Laya, Kev, Decider, Bespoke Nimble, Von, CLM and others publish their weights, many under Apache-2.0. The list of open alternatives to Jev, with licences and which ones speak its API, is at systemonemodels.tech/jev-alternatives.
pip install systemonemodels, then systemone pull namespace/name downloads an open model with every file checked against its SHA-256. Hosted ones are called through their provider’s API; every model page has a playground that builds the request.
Read on: Decision model, defined · Typed questions · Calibrated confidence · Why decision models need their own registry