Jev API on TokenRa

What is Jev?

Jev is a structured evaluation model built by TypeSafe AI and announced on September 15, 2026 by Diogo Almeida. TypeSafe describes it as a "System One" model — a reference to the fast, intuitive decision‑making described in Daniel Kahneman's Thinking, Fast and Slow. While chat models (System Two) are good at writing prose and reasoning step‑by‑step, Jev is designed to make one decision, fast, and give you a number you can trust.

It is currently in early access. TokenRa provides a unified API endpoint so you can start building with Jev today without managing a separate TypeSafe account.

Jev at a glance

How Jev differs from chat models

Most AI models are chat models: you send a list of messages and they return the next token in a sequence. Jev works differently:

• Input: a state (the text to evaluate) plus one or more questions (Noul, Choice, or Score).
• Output: a typed decision object — not a sentence. For a Choice question you get per‑option probabilities; for a Score question you get a number; for a Noul question you get true or false.
• Confidence: every answer comes with a calibrated confidence score, so your code can decide whether to trust the result or fall back to a human review step.

This makes Jev a better fit than a chat model for tasks like content moderation, intent classification, lead scoring, NLU routing, guard‑rails, and automated QA — anywhere you need a fast, reliable, typed decision rather than a paragraph of prose.

Getting a Jev API key on TokenRa

1. Sign up at tokenra.io/sign-up — no platform subscription required.
2. Create an API key in your TokenRa dashboard. Give it a descriptive label so you can track Jev usage separately.
3. Send your first Jev request using the example below. Set the Authorization header to Bearer YOUR_TOKENRA_KEY and the endpoint to https://api.tokenra.io/v1/models/jev-latest.
4. Pay per use — you are billed per 1M input tokens at TokenRa's rate. Output tokens are free. See the pricing section below.

Request and response example

Jev does not use the OpenAI chat format. Its request body uses state plus questions. Below is a working example you can copy and run.

cURL

curl https://tokenra.io/v1/decisions \
  -H "Authorization: Bearer YOUR_TOKENRA_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "jev-latest",
    "state": "Help! My payouts have been failing for 3 days.",
    "questions": {
      "is_urgent": {
        "type": "noul",
        "instructions": "Does this message convey urgency?",
        "criteria": {
          "true": "Explicitly time-sensitive",
          "false": "No urgency expressed"
        }
      },
      "department": {
        "type": "choice",
        "instructions": "Which team should handle this?",
        "criteria": {
          "billing": "Payments, invoicing, refunds",
          "technical": "Bugs, outages, integrations",
          "sales": "Pricing, upgrades, new accounts"
        }
      },
      "frustration": {
        "type": "score",
        "instructions": "How frustrated is the customer?",
        "criteria": ["Calm", "Frustrated", "Very angry"]
      }
    }
  }'

JavaScript (fetch)

const response = await fetch(
  "https://tokenra.io/v1/decisions",
  {
    method: "POST",
    headers: {
      Authorization: "Bearer YOUR_TOKENRA_KEY",
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      model: "jev-latest",
      state: "Help! My payouts have been failing for 3 days.",
      questions: {
        is_urgent: {
          type: "noul",
          instructions: "Does this message convey urgency?",
          criteria: {
            true: "Explicitly time-sensitive",
            false: "No urgency expressed"
          }
        },
        department: {
          type: "choice",
          instructions: "Which team should handle this?",
          criteria: {
            billing: "Payments, invoicing, refunds",
            technical: "Bugs, outages, integrations",
            sales: "Pricing, upgrades, new accounts"
          }
        },
        frustration: {
          type: "score",
          instructions: "How frustrated is the customer?",
          criteria: ["Calm", "Frustrated", "Very angry"]
        }
      }
    }),
  }
);

const data = await response.json();
// data.answers.is_urgent.noul → 0.95 (probability)
// data.answers.department.choice → "billing"
// data.answers.department.probabilities → { billing: 0.89, technical: 0.11, sales: 0 }
// data.answers.frustration.score → 1.04
// data.answers.frustration.confidence → 0.93

Expected response shape

{
  "model": "typesafe/jev-1.13-20260917",
  "answers": {
    "department": {
      "type": "choice",
      "choice": "billing",
      "probabilities": {
        "billing": 0.89,
        "technical": 0.11,
        "sales": 0
      },
      "confidence": 0.83
    },
    "frustration": {
      "type": "score",
      "score": 1.04,
      "legend": {
        "0": "Calm",
        "1": "Frustrated",
        "2": "Very angry"
      },
      "probabilities": {
        "0": 0,
        "1": 0.95,
        "2": 0.05
      },
      "confidence": 0.93
    },
    "is_urgent": {
      "type": "noul",
      "noul": 0.95
    }
  },
  "usage": {
    "input_tokens": 427,
    "output_tokens": 73,
    "cost": 0.000017934
  },
  "id": "gen-dec-1789842178-QJqpDPsaQ5UlheoIoMfo",
  "provider": "TypeSafe"
}

Jev API pricing on TokenRa

Jev is billed per 1M input tokens at TokenRa's rate. Output tokens are free. There is no platform subscription — you only pay for what you use.

See all models and pricing or read the billing FAQ for details on how TokenRa counts tokens and handles overage.

When to use Jev — and when a chat model is the better call

Jev is not a replacement for chat models; it is a complement. Here is a quick decision guide:

Use Jev when you need to …

• Classify user intent from a support message
• Score lead quality from a CRM note
• Detect toxicity, spam, or policy violations
• Route a request to the right handler in a multi‑agent pipeline
• Gate an LLM output — "should I show this to the user?"
• Run a fast guard‑rail check before an expensive chat call

Use a chat model when you need to …

• Write an email, blog post, or product description
• Hold a multi‑turn conversation
• Summarize a long document
• Generate code or creative content

The two work best together: Jev decides, Omen‑Alpha (or another chat model) writes.

FAQ

Is Jev an AI model from TypeSafe?

Yes. Jev is TypeSafe AI's first System One model, released in September 2026. It returns typed decisions — Noul, Choice, and Score — each with a calibrated confidence value, instead of free‑form text.

Does Jev write text?

No. Jev does not generate prose, chat replies, or code. It evaluates a state against structured questions and returns typed answers with probabilities. For text generation, use a chat model like Omen‑Alpha or Union‑Alpha.

Is Jev free?

Jev is available on TokenRa on a pay‑per‑use basis: you are charged per 1M input tokens; output tokens are free. TypeSafe may offer a free tier directly — check typesafe.ai for their current terms.

Is Jev compatible with the OpenAI chat format?

No. Jev expects a state field plus a questions array — not a messages array. Copy the request examples on this page for the correct shape. The endpoint returns typed answers, not chat completion chunks.

Jev vs Omen‑Alpha — which one should I use?

Use Jev for fast, typed decisions (classification, scoring, routing, guard‑rails). Use Omen‑Alpha for text generation, conversation, or creative output. They are complementary: many production pipelines call Jev for evaluation and a chat model for generation.

Related models & docs

• All models on TokenRa — compare pricing, context windows, and use‑cases
• Ox‑Alpha API — fast, cost‑effective chat model
• Union‑Alpha API — high‑capacity chat model for complex tasks
• TokenRa API docs — authentication, rate limits, error codes
• Jev API deep‑dive guide — third‑party tutorials, use‑cases, and comparison articles