Jev: an AI model that returns decisions, not text

What is Jev?

TypeSafe AI is a San Francisco company founded in 2024. Jev is its first public "System One" model, built for decisions inside software rather than for chat. Limited early access began on 15 September 2026, alongside a $40 million seed round.

Jev is proprietary. TypeSafe hasn't published its weights, its architecture or a technical paper, so what we know comes from the company's own description.

How does Jev work?

Each request has a state, which is your text or JSON, and one or more questions about it. Every question is evaluated in parallel in a single call. There are three question types:

Type	What it does	Returns
Choice	Picks one option from a list you define	The choice, probabilities and confidence
Score	Rates the state against levels you define	A score, probabilities and confidence
Noul	Judges whether a statement is true	A value from 0 to 1

Here is a trimmed example request, from the quickstart, to POST https://api.typesafe.ai/v1/systemone:

{
  "state": "Your text to evaluate goes here.",
  "model": "jev-latest",
  "questions": {
    "department": {
      "type": "choice",
      "instructions": "Which team should handle this",
      "criteria": {
        "billing": "Payment or subscription issues",
        "technical": "Bugs or integration problems"
      }
    },
    "is_urgent": {
      "type": "noul",
      "instructions": "The message conveys urgency"
    }
  }
}

The answers come back keyed by the question names you chose:

{
  "answers": {
    "department": {
      "type": "choice",
      "choice": "technical",
      "confidence": 1.0,
      "probabilities": { "technical": 1.0, "billing": 0.0 }
    },
    "is_urgent": { "type": "noul", "noul": 0.99 }
  }
}

Because the possible answers are fixed in advance, the output always fits the shape your code expects. TypeSafe's advice is to ask one specific question per factor and combine the results in code, instead of asking one broad question.

Why it's interesting

Most AI in business software is really a decision: which team gets this message, is this lead worth a call, does this draft meet the checklist. A chat model can answer those, but you then have to parse its prose and hope it stays consistent. Jev returns a value and a confidence, so your software can act when confidence is high and hand the case to a person when it isn't.

Numbers to treat carefully

• TypeSafe lists input tokens at $42 per billion.
• It claims response times of 70 to 500 milliseconds, and results up to 193.6 times faster and 444.6 times cheaper than frontier language models on its own workflows.
• Those workflows were built by TypeSafe's team, and the company says real-world gains will likely be lower. We haven't benchmarked Jev ourselves.

How to get started with the Jev API

Create an API key in the TypeSafe console and store it as TYPESAFE_API_KEY. Official SDKs exist for Python and for JavaScript or TypeScript:

pip install typesafe-sdk
npm install @typesafe-ai/sdk

Access is limited early access, so check the documentation for current availability.

Where we think it could fit

These are ideas, not shipped work. Sorting website enquiries into the right category with an urgency score. Checking a content draft against a checklist before a person reviews it. Deciding whether an automation step should run on its own or wait for approval. All three suit a model that answers with a number and a confidence.

Frequently asked questions

What is Jev AI?

Jev is a proprietary AI model from TypeSafe AI and its first public System One model. It answers structured questions with typed values, probabilities and confidence scores instead of writing text.

How is Jev different from a chat model?

A chat model writes sentences that your code has to parse. Jev returns a choice, a score or a true-or-false value in a fixed format, so software can act on the answer directly.

How much does Jev cost?

TypeSafe lists input tokens at $42 per billion. Check typesafe.ai for current pricing, as access is in limited early access.

How do I use the Jev API?

Create an API key in the TypeSafe console, store it as TYPESAFE_API_KEY, then send a state and your questions to the System One endpoint. SDKs are available for Python and for JavaScript and TypeScript.

Is Jev available to everyone?

Not yet. Limited early access began on 15 September 2026, so check the TypeSafe documentation for current availability.