A support ticket enters a queue at 9:14 a.m. The message says a payment connection has failed for three days. Ordinary software sees a string. Jev reads the same text, classifies the problem, scores the urgency, and returns numbers that code can use.

That narrow function explains Jev AI. TypeSafe AI released Jev in September 2026 as its first System One model. It does not chat, write reports, or produce open ended answers. It makes bounded decisions inside software.

The idea matters because more teams now place AI inside repeatable workflows. The 2025 Stanford AI Index reported that the cost of querying a model at roughly GPT 3.5 performance fell from $20 to $0.07 per million tokens between November 2022 and October 2024. That was a reduction of more than 280 times. Jev follows the same cost trend, but it removes text generation from the job.

What is Jev AI?

A model that returns decisions, not prose

Jev is a hosted AI model made by TypeSafe AI. You send it a state, which can be text or structured text data, plus one or more focused questions. It returns typed answers such as a selected category, a score, or a probability.

TypeSafe calls this class of model System One. The name refers to fast, immediate judgment rather than slow, multi step reasoning. In the company’s words, founder Diogo Almeida describes Jev as a “frontier intelligence function call: unstructured state in, typed probabilistic decisions out.” The official launch post also says TypeSafe trained the model with Reinforcement Learning for Calibrated Decisions, or RLCD.

This distinction separates Jev from a general large language model. An LLM can draft an email, explain code, or produce a long answer. Jev cannot. It selects from answer shapes that you define before the request. This makes the response easier for software to validate and act on.

You can find more reporting on automation in Ufaq’s AI section and broader product coverage in the Technology section.

How the Jev System One model works

Choice, Score, and Noul questions

The Jev System One API accepts three question types.

  • Choice selects one option from a list. A support system can choose billing, technical, or sales.
  • Score places the state on an ordered scale. A moderation system can rate severity from low to high.
  • Noul returns a probability from 0 to 1 for a yes or no statement. A security layer can estimate whether a prompt contains an injection attempt.

Choice and Score answers include a probability distribution and a confidence value. Noul returns the probability that the statement is true. You can ask several independent questions in one request. Jev evaluates them against the same state in parallel.

That design gives your code control. A high confidence result can move forward. A medium confidence result can request more information. A low confidence result can go to a human reviewer. The model supplies judgment. Your software controls the action.

The phrase System One Jev sometimes appears in searches, but it refers to the same product. System One is the model class. Jev is TypeSafe AI’s first public model in that class.

How to use Jev

Send state and focused questions to one endpoint

Start in the TypeSafe playground or create an API key in the TypeSafe console. The official Jev quick start uses the endpoint shown below.

POST https://api.typesafe.ai/v1/systemone

Send your key in the Authorization header. Then provide the state, a model name, and your questions. This sample routes a customer message and checks its urgency.

{
  "state": "My card was charged twice and I need help today.",
  "model": "jev-latest",
  "questions": {
    "department": {
      "type": "choice",
      "instructions": "Which team should handle this message?",
      "criteria": {
        "billing": "Payment or subscription issue",
        "technical": "Product failure or software bug",
        "sales": "Purchase or plan question"
      }
    },
    "urgent": {
      "type": "noul",
      "instructions": "The customer needs action today"
    }
  }
}

Write one clear judgment in each question. Define every option in plain language. Keep dates, counting, and arithmetic in normal code. If the decision has several parts, ask several atomic questions and combine the results after the API responds.

You can place Jev inside a ticket router, an agent guardrail, a document review process, or a model selection layer. For more implementation material, check Ufaq’s Guides section. If you want to compare Jev with a general chat model, read Ufaq’s ChatGPT 5 overview.

Jev pricing, models, and current limits

Jev price in September 2026

The official Jev models reference lists Jev 1.13 at $0.042 per million input tokens, or $42 per billion input tokens. Output tokens are free. TypeSafe charges for the text sent into the model.

The current versioned model ID is jev-1.13.0. The aliases jev-latest and jev-preview both point to that version as of September 30, 2026. TypeSafe can move an alias when it releases a new version. Pin the full model ID if you need stable behavior during testing.

Jev supports a 64,000 token request budget. The state plus the longest question has a separate 32,000 token limit. The published standard rate limits are 100,000 tokens per second and 40 requests per second. TypeSafe says those limits can change while it expands capacity.

Company tests report response times from 70 to 500 milliseconds. TypeSafe also reports that Jev was 193.6 times faster and 444.6 times cheaper than the compared LLM workflows. These are company benchmarks, not a universal result. Measure speed, accuracy, and cost on your own data before deployment.

Is Jev free or open source?

The query “Jev free” needs a precise answer. Output tokens are free, but TypeSafe charges for input tokens. The playground can help you inspect the request format, but you should confirm current account access and billing in the console.

Jev is not an open source model. TypeSafe provides a hosted API and public SDKs, but it has not released the Jev model weights. Search phrases such as “Jev open source” and “open source Jev” can lead to community projects built around the service. Those projects do not make the underlying model open source.

Where Jev fits and where it does not

Good uses for fast, bounded judgment

Jev fits tasks where software must choose, score, route, or verify. Common examples include customer support routing, content moderation, document classification, search ranking, citation checks, model routing, prompt injection detection, and review queues.

The model also suits high volume work. One state can carry several questions, and Jev evaluates those questions in parallel. This lets a system check topic, urgency, policy risk, and routing in one request.

Jev does not remove the need for ordinary code. Code should still handle permissions, calculations, database writes, exact date comparisons, and final business rules. A human should review uncertain results when an error can harm a customer or create legal or financial risk.

Known limits in Jev 1.13

TypeSafe’s Jev 1.13 limitations page lists several weak areas. The model can read instructions too literally. It does not count reliably. It struggles with numeric precision, date comparison, complex indirection, adversarial text, and long states filled with irrelevant details.

Jev also accepts text only. Convert images, audio, and video into useful text or structured fields before sending them. English is its strongest language. Test other languages on representative data before you automate decisions.

The practical rule is simple. Use Jev for semantic judgment inside a controlled workflow. Use code for exact operations. Use a generative model when you need new text, plans, explanations, or creative output.

Jev AI FAQ

Common questions about TypeSafe AI and Jev

What is Jev AI?
Jev AI is TypeSafe AI’s first System One model. It reads a state, answers typed questions, and returns choices, scores, probabilities, and confidence values for software.

Is Jev an LLM?
No. TypeSafe presents Jev as a System One decision model, not a large language model. It does not generate open ended text.

What Jev models are available?
The current model is Jev 1.13, identified as jev-1.13.0. You can also call the jev-latest or jev-preview alias.

What is the Jev pricing model?
Jev pricing is usage based. TypeSafe lists input at $0.042 per million tokens. Output tokens are free. Check the official model page before setting a production budget.

How do you use Jev?
Send a POST request to the System One endpoint with your state, model ID, and one or more Choice, Score, or Noul questions. Your code then acts on the typed result.

Do “jev ia,” “ia jev,” and “jev ja” refer to the same tool?
These search phrases usually point to Jev AI from TypeSafe. “IA” is also the abbreviation for artificial intelligence in several languages. Check the domain and product description because JEV can refer to unrelated subjects.