Jev is a System One decision model from TypeSafe AI. Given a blob of state and a question with fixed answer choices, it returns a typed, probabilistic answer in 70 to 500 milliseconds, at roughly $0.0004 per decision.

What does JEV stand for?

JEV is the product name TypeSafe AI uses for its System One decision model. The three primitives are Choice, Score, and Noul: a fixed selection, a numeric rating, and a calibrated yes/no probability.

How is Jev different from ChatGPT?

Jev answers fixed-choice questions with typed output and no generated prose. ChatGPT generates free-form text. Jev is for high-volume structured judgment; an LLM is for writing and open-ended reasoning.

Can Jev generate content?

No. Jev does not write articles or descriptions. It returns selections, scores, and probabilities. Use it to classify, sort, filter, and judge; use a generative model for the writing.

Yes, for structured judgment: keyword intent classification, freshness scoring, internal link mapping, cannibalization flags, and page element audits. See the jev for SEO guide for a step-by-step workflow.

GLOSSARY · 2026-09-22 · READ TIME 3 MIN

What is Jev?

Jev is a System One decision model from TypeSafe AI. Give it a blob of state and a question with fixed answer choices, and it returns a typed, probabilistic answer in 70 to 500 milliseconds, at roughly $0.0004 per decision, up to 250k tokens per second. It decides instead of writing.

The three primitives

Primitive	What it returns	Use for
Choice	One option from a fixed list, plus confidence	Intent classification, funnel stage, keep-or-change verdicts
Score	A numeric rating, plus confidence	Freshness scoring, competitor answer depth, GEO citation likelihood
Noul	A calibrated yes/no probability	Should this page get a link, is this title duplicate, is this safe to run

Every answer carries a confidence value. The convention in the ecosystem: for destructive or critical operations, require confidence of at least 0.80 before acting.

Jev versus a generative LLM

Axis	Jev	ChatGPT / Claude
Output	Typed, structured, probabilistic	Free-form text
Cost per decision	About $0.0004	Tens to hundreds of tokens billed
Speed	70 to 500ms per question	Seconds per generation
Job	Sort, classify, score, judge	Write, explain, reason open-ended
Strengths	High-volume repeatable judgment	Novel creative output

What Jev does not do

It does not generate prose: no articles, no meta descriptions from scratch, no paragraph explanations.
It does not crawl the web or fetch URLs. A crawler or a client like jev-seo collects state and feeds it in.
It is not a calculator. Arithmetic, counting, and date math stay in host code; Jev handles semantic judgment only.
It is not a chat model. You batch independent questions against one state snapshot, one request, and read the answers.

Where Jev fits in SEO

The full workflow is on the jev for SEO guide. In one line: a client collects state (SERP results, Search Console rows, local page files) and batches Jev questions against it, turning a $250 per-client audit into a fraction of a cent of judgment calls, with a typed transcript instead of a prose report.

Frequently asked questions

Is Jev open source? Jev is a hosted decision API by TypeSafe AI. The client ecosystem around it, including jev-seo, is MIT-licensed open source.

How much does Jev cost? Roughly $0.0004 per decision at list price, about $0.042 per million input tokens, with output tokens free. Hard limits per request: 32k tokens for state plus the longest question, 64k total context.

Can Jev replace an LLM in my stack? No. It replaces the expensive repetitive judgment calls. Keep a generative model for writing and pair them: Jev sorts and scores, the LLM writes.

NEXT: jev for SEO guide → · home