What is Jev? TypeSafe’s System One evaluation model explained

Jev is TypeSafe AI’s first public System One model: a probabilistic decision engine for software. You send it shared state and typed questions. It returns structured answers with probabilities, not chat paragraphs you have to parse.

TypeSafe positions System One models after Kahneman’s fast thinking: optimized for quick, calibrated decisions inside code. Jev is named after economist William Stanley Jevons. The bet is that cheaper intelligence unlocks far more automation, the way cheaper energy unlocked more industry.

On Vercel AI Gateway the model id is typesafe-ai/jev. Answers are structured values rather than long generated text, so you get typed decisions without parsing chat output.

A language model is built to generate strings. That flexibility is powerful for chat and drafting, but software that needs a reliable branch still has to parse, validate, and hope the model did not invent a new shape.

Jev flips that contract. Possible answers are declared up front. The model returns type-safe values with calibrated probabilities and confidence. TypeSafe reports that schema matching is guaranteed, so the kind of hallucinated tool call that breaks an agent loop is out of scope by design.

Sampling is parallel rather than token-by-token. TypeSafe’s published claims put end-to-end latency in roughly the 70ms to 500ms range for System One shaped queries, and they argue Jev can be tens to hundreds of times faster and cheaper than frontier LLMs on those decision workloads.

Boolean (Yes / No) returns a probability from 0 to 1 that the statement is true. Use it for guardrails, verification, and continue-or-stop checks.

Choice picks one option from a set you name, with a probability mass over the alternatives. Use it for routing tickets, picking a tool, or selecting the next agent step.

Score rates state against an ordered rubric you define. Use it for urgency, risk, quality, or priority before an action runs.

One request can evaluate many questions over the same state in parallel, which is how Jev fits clean workflows instead of a single megaprompt.

Vercel reported that within 24 hours of launching on AI Gateway, Jev reached more than twice as many paid teams as any previous model launch, becoming the fastest-adopted model in Gateway history.

Common patterns: choose an agent’s next tool, decide whether a workflow should continue, retry, ask a human, or stop, score urgency before an action, and verify or guardrail other model outputs when confidence is low.

That is the same decision layer agentic products need. On DAKAEi, agents already connect to the apps you work in. Evaluation models like Jev are how those agents can make fast, typed calls without turning every branch into another free-form chat turn.

If you want to feel the product before wiring APIs, open the free playground at /try/jev. Ask a yes-or-no question, force a choice between options, or score something on a rubric. The page calls Jev through AI Gateway from DAKAEi.

When you are ready to put decisions next to real work, open DAKAEi chat, connect your apps, and build agents that act inside email, calendar, documents, and code. Jev is for the decision. DAKAEi is for the work that follows.