Primary Entity:jev-1.13.0(jev-latest)

What is Jev AI?

Jev is a System One decision model developed by TypeSafe AI. Unlike generative large language models (LLMs) that produce autoregressive text tokens, Jev evaluates structured software state and returns typed decisions and probabilities across predefined schemas.

Jev does not chat, write prose, execute agentic loops, or generate free-form text. TypeSafe documents a dual-budget context constraint: a 64,000-token total request budget across state plus all questions combined, with a 32,000-token cap for state plus the single longest question (OpenRouter lists Jev 1.13 with a 32,000-token context window). Evaluations run using three native primitives—Choice, Score, and Noul—at published rates of $0.042 per million input tokens and $0.00 for output tokens.

Verified Technical Specifications

Source: TypeSafe Documentation (Sep 2026)

Model Identifierjev-1.13.0Alias: jev-latest

Model ClassificationSystem One Decision ModelNon-autoregressive decision model (trained via RLCD)

Context Window Limits64,000 / 32,000 dual-budgetTypeSafe: 64k total, 32k state + longest Q (OpenRouter: 32k context listing)

Published Token Pricing$0.042 / 1M input · $0.00 outputFixed rate via TypeSafe AI & OpenRouter

Evaluation PrimitivesChoice · Score · NoulDiscrete options (≤255), ordered rubrics (2–10 levels), binary truth (0.0–1.0)

Primary API RoutePOST /v1/systemoneSupported natively by TypeSafe AI and OpenRouter

What do you need to do with Jev?

Select your specific implementation task to jump directly to verified technical references, schema specifications, and runnable code:

API Access & Quickstarts →Code

Connect via Direct TypeSafe API or OpenRouter. Runnable examples in cURL, Python (typesafe_sdk), and TypeScript (@typesafe-ai/sdk).

Decision Primitives Guide →Schema

Choose between Choice (closed discrete sets ≤255), Score (ordered rubrics 2–10), and Noul (binary truth probability).

Jev vs. Generative LLMs →Architecture

Compare System One decision networks with autoregressive generative models. Architectural trade-offs, determinism, and hybrid pipelines.

Pricing & Cost Calculator →$0.042/M

Published rates across verified providers. Calculate exact token expenditures from your projected input token volumes.

Creator & Claims Ledger →Evidence

TypeSafe AI founding team (Diogo Almeida, Sasha Sheng, Erik Gafni), RLCD training claims, verified observable facts, and unverified benchmarks.

Release Changelog →Ledger

Dated timeline of official model version releases, platform integrations (OpenRouter, Vercel AI Gateway), and schema adjustments.

Minimal Request Example

The standard TypeSafe System One payload passes an application state block and one or more evaluation questions:

Bash / cURLbash

curl -X POST "https://api.typesafe.ai/v1/systemone" \
  -H "Authorization: Bearer $TYPESAFE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "jev-latest",
    "state": "User has attempted payment 3 times with card ending in 4022. IP country matches billing address. Device fingerprint unchanged. Previous transaction 12 minutes ago was approved.",
    "questions": [
      {
        "id": "fraud_action",
        "type": "choice",
        "prompt": "Recommended transaction handling action",
        "choices": ["approve", "require_mfa", "manual_review", "hard_decline"]
      }
    ]
  }'

Official endpoint: https://api.typesafe.ai/v1/systemoneView full Python & TypeScript SDK guides →