---
title: "What is Jev: typed choice, score, and yes/no answers from text or JSON | system-one.dev"
description: "Jev is TypeSafe's decision model on system-one.dev. Send text or JSON and get a choice, a score, or a yes/no probability—without generating prose."
canonical: "https://system-one.dev/en/models/jev"
language: "en"
---

# Jev

TypeSafe's decision model. It reads text and JSON and returns a choice, a score, or a yes/no answer.

Jev is TypeSafe's decision model, provided by system-one.dev. Give it some text or JSON and a few questions; it picks an option, gives a score, or answers a yes/no question. Use the model name returned by GET /v1/models.


## Model info

- Provided by: system-one.dev
- 1 credit per question per 4KB
- Up to 32 questions · 64KB per request
- Inputs: text and JSON
- Pick one: up to 255 options; Score: 2–10 levels
- Released 2026-09-18

## Uses

- **Tool routing**: Pick one candidate tool, or decide that none of them fit. See [Agent tool routing](https://system-one.dev/en/blog/agent-tool-routing).
- **Model routing**: First judge how hard the request is, then send it to a small or large model. See [Model routing](https://system-one.dev/en/blog/model-routing).
- **First-pass evaluation**: Use Score instead of an LLM as judge. See [Using Score instead of an LLM as judge](https://system-one.dev/en/blog/llm-as-a-judge-alternative).
- **Completion checks and guardrails**: Use a yes/no probability to stop, allow, or hand off to a person. See [When an agent should stop](https://system-one.dev/en/blog/agent-stop-condition) and [Agent guardrails](https://system-one.dev/en/blog/agent-guardrails).
- **Try it**: [Tweet moderation](https://system-one.dev/en/demo/tweet-moderation), [Phishing-email](https://system-one.dev/en/demo/phishing-email) (sample data; sign in required).

## Frequently asked questions

### Is Jev an LLM?

No. It does not generate text. It only returns a choice, a score, or a probability. Use a generative model when you need written content.

### Does Jev support Chinese?

It accepts Chinese input. TypeSafe's docs say English works best. Other languages work, but accuracy is lower than English, so test with your own data first.

### How is Jev billed on system-one.dev?

By credits. Each question is billed in request-byte blocks. See Model info on this page for the rate. Details: [Pricing](https://system-one.dev/en/pricing).

### Do versions change?

Yes. TypeSafe's official alias `jev-latest` moves when a new version ships. If you tuned thresholds on one version, re-check them after a change. On this site you can pick a full model ID; if you omit it, the default model is used.

### Can Jev see images?

No. TypeSafe's docs say Jev accepts text input only.

### Can I fine-tune it?

No. TypeSafe's docs say they do not fine-tune on customer data, and every account shares the same weights. To fit your task, put material in state and write the rules in the question's instructions and criteria.

## Official sources

- [TypeSafe model docs](https://docs.typesafe.ai/models)
- [TypeSafe System One concepts](https://docs.typesafe.ai/concepts/system-one)
- [TypeSafe launch blog](https://typesafe.ai/blog/introducing-system-one-models-and-jev)
- [On this site: decision primitives](https://docs.system-one.dev/en/docs/primitives)

Last updated 2026-10-12

- [Open GET /v1/models](https://system-one.dev/v1/models)

- [Developer docs](https://docs.system-one.dev/en/docs/api#developer-details)

- [Try live (sign in required)](https://system-one.dev/en/demo/tweet-moderation)

- [All models](https://system-one.dev/en/models)