# Jev FAQ

> Jev is a typed decision model from TypeSafe AI; JevSkills is an independent place to discover practical projects, skills, videos and guides around it.

## What is Jev?

Jev is TypeSafe AI's typed decision model. An application supplies structured state and receives a constrained decision instead of an open-ended paragraph.

## Is Jev an LLM?

TypeSafe AI presents Jev as a System One model, not a general-purpose large language model. It is designed for narrow, typed decisions rather than conversational prose.

## Why does Jev not generate paragraphs?

Its interface is intended to return a bounded decision that software can inspect and validate. A host application can explain that decision to a person when it needs to.

## What are Choice, Score, and Noul?

Choice selects one allowed option. Score returns a bounded numeric assessment. Noul is a yes-or-no decision with a probability. The exact schema is defined by the application using Jev.

## How is Jev different from GPT or Claude?

GPT and Claude are general-purpose language models that commonly generate text. Jev is positioned for a smaller decision boundary with typed output; the host still owns validation, policy and actions.

## What can I build with Jev?

Useful starting points include routing, triage, scoring, eligibility checks and human-review decisions—where an application needs a clear, bounded result.

## How do I use Jev with Claude Code?

Use Claude Code to help implement an application boundary around Jev: keep credentials server-side, send structured inputs and validate the returned type before acting. Check TypeSafe AI's current documentation for supported SDKs and setup.

## How do I use Jev with Codex?

Codex can help build the same integration pattern: model the input and allowed output, call Jev from a secure server boundary, then validate before a downstream action. Follow the current provider documentation for the supported client.

## How do I use Jev with Cursor?

Cursor can assist with implementing a typed Jev boundary in your codebase. Treat it as a coding environment, not a replacement for provider documentation, application validation or secure secret handling.

## What is Jev useful for?

Jev is useful when a product needs one understandable, structured decision instead of a long generated answer—for example, where to route work or whether a case needs review.

## Is Jev deterministic?

A typed response is not by itself a guarantee of mathematical determinism. Validate the behavior you need against the current provider documentation, model version and your own test cases.

## Does Jev hallucinate?

Bounded output does not make a model infallible. Treat Jev results as inputs to application logic, validate important decisions and keep human review for consequential cases.

## How fast is Jev?

A TypeSafe-recorded demo reported Jev at 0.114 seconds versus GPT-5.6 Terra at 8.566 seconds for one 27-question vendor-run example. That is context for a specific demo, not a universal benchmark.

## What does Jev cost?

Pricing can change. Check TypeSafe AI's current pricing and documentation before planning a production integration.

## Where can I find Jev projects and skills?

Explore JevSkills for source-screened builds, skills, GitHub projects, videos and learning guides. A sign-in is required only for protected details and high-intent actions.

## What is JevSkills?

JevSkills is an independent community resource for discovering practical Jev projects, skills, videos and guides. It labels evidence so visitors can distinguish source pointers from tested work.

## Is JevSkills affiliated with TypeSafe AI?

No. JevSkills is an independent community resource and is not affiliated with TypeSafe AI.

## Sources and updates

Updated September 22, 2026. Product descriptions and the benchmark qualification above are based on TypeSafe AI's published Jev introduction; availability, pricing and SDK support should always be checked in current provider documentation.

Source: [TypeSafe AI: Introducing System One Models and Jev](https://typesafe.ai/blog/introducing-system-one-models-and-jev)