Jev Model

It answers questions. It doesn’t write paragraphs.

The site calls Jev Model a “System One” model, after the fast, intuitive mode of thinking, as opposed to the slow, deliberate one. In practice that means small, repeated judgments made inside software: which team should this ticket go to, how angry is this customer, is this tool call about to delete something. Every question has to be one of three types:

• Choice picks one answer from options you define (up to 255 of them) and returns the winner, a probability for every option and a confidence score.
• Score rates the input on an ordered scale of 2 to 10 levels you describe, such as “calm, frustrated, very angry”. The returned score is probability-weighted, so it can land between two levels.
• Noul is the site’s name for a yes/no judgment. It returns a single number from 0 to 1: the probability that the answer is yes.

Because you define the answer space up front, the output can go straight into an if statement, a queue or a database column. You don’t need a second prompt to clean it up.

One state, up to eight questions, one request

Each request carries a shared “state”, the context every question reads. The documentation allows three forms: plain text, a JSON object (for example a ticket, an order and a policy together), or an array of text items. The questions are then answered in parallel against that same state, so a single call can classify a support ticket, score its urgency and decide whether a person should see it, instead of chaining three separate calls.

The limits are stated plainly in the docs. Images, audio and video are not supported yet. The whole thing runs through one endpoint, POST /v1/systemone, with a bearer API key. The site also offers an “Agent Skill” that teaches coding agents such as Claude Code, Codex and Cursor to call the API for bounded decisions while the application keeps control of the final action.

The playground ships with fifteen ready-made scenarios, from support triage and lead scoring to content moderation, résumé screening, model routing and a guardrail for risky tool calls. They are the quickest way to see what a well-scoped question looks like.

The probability is the real product

Plenty of tools can put a label on a message. What Jev Model adds is a number you can set a threshold against: route automatically above 0.8, send to a person below it. The homepage’s example code does exactly that, escalating a ticket only when the urgency probability clears a bar.

To its credit, the documentation doesn’t oversell this. It says the probabilities are “signals for automation, not a guarantee of business accuracy,” and tells developers to use higher thresholds or human review for high-risk actions. We agree, and we’d add that the thresholds only mean something once you’ve tested them against your own data. The homepage also quotes a typical latency of about 120 milliseconds per decision. That is the vendor’s figure, and we couldn’t measure it ourselves.

Free to try, paid to ship

The playground is free with unlimited runs, but it needs an account. When we sent a request without signing in, the service answered “Please sign in to use the playground.” At the time of this review the pricing page listed three one-time access plans, each with unlimited usage for its period:

• Starter, about $9.90 for 7 days: one workspace, one concurrent request, standard speed, API key management and email support.
• Pro, about $99 for 30 days: unlimited workspaces, three concurrent requests, a faster lane, request history and priority support.
• Enterprise, about $999 for 365 days: ten concurrent requests, team workspaces and custom integration support.

The page says one-time plans don’t auto-renew, which is useful if you only want to test a single workflow for a week. “Unlimited” refers to the number of requests. The real cap is concurrency, and one request at a time on Starter is fine for a prototype but won’t carry production traffic.

No Arabic interface, and Arabic input is untested

At review time the site was available in English and Simplified Chinese only. There is no Arabic interface, and the Agent Skill’s guidance language can be set to English or Chinese, nothing else. Nowhere in the documentation is Arabic mentioned, either as a supported input language or as a tested one.

Because the state is free text, nothing technically stops you from sending an Arabic support ticket or review. We couldn’t confirm how well the model classifies it, and the site makes no promise either way. If Arabic content is your main use case, run a batch of real Arabic examples through the playground and compare the probabilities with the English equivalents before paying for API access.

What the site doesn’t tell you

The “About” page describes the product and gives a support email, but it doesn’t name the company, the team or the country behind it, and we found no launch date or funding information. The docs don’t say what the model is built on either. One line in the API reference calls jev-latest “TypeSafe’s flagship model,” while the site footer states that Jev AI is not affiliated with TypeSafe. We can’t resolve that from the public pages. It’s worth asking the vendor before you send customer data through the API.

Who it’s for, and who should skip it

Jev Model suits developers and small product teams who already have a stream of text to sort: support tickets, form submissions, user posts, agent tool calls. They want those decisions as clean, typed fields with a confidence number, not as prose. It’s also a reasonable fit for anyone building AI agents who wants a cheap, fast check before a risky action, or a router that sends easy tasks to a smaller model.

It’s the wrong tool if you need generated text, summaries or images, if your inputs are images or audio, or if you don’t write code at all: without the API, it’s a demo. Teams that handle mostly Arabic content, or that need a named vendor and a clear data-processing agreement, should do their own testing and due diligence first.