What is Jev?

Jev is a decision model built by TypeSafe AI, launched in September 2026. You don't chat with it and it doesn't write anything. You ask it small, closed questions ("is this email spam?", "which queue does this ticket belong in?", "safe to proceed?") and it returns a structured answer, in milliseconds, for a fraction of a cent.

That narrow job description is the entire product, and it turned out to unlock a lot: within days of launch, builders had shipped hundreds of working systems on it, which is why this directory exists. (Disambiguation for the search engines and the confused: this page is about Jev the AI model, not FaZe Jev the YouTuber.)

How it works, in one mental model

Think of Jev as a judge, not an author. A chat model like GPT or Claude takes a request and composes a response. Jev takes a question with a fixed answer set (a label from your list, a yes/no, a score, a choice) and rules. The output is natively structured: there is no prose to parse, no JSON to repair, no essay to strip. You frame the task as questions; it answers them, extremely fast, at very high volume.

Framing tasks as questions is a real skill with its own name and its own page: What is Jev Engineering.

What it's being used for

The catalog's eight categories reduce to one sentence: anywhere software needs many small judgments, fast and cheap. The reported reference points, receipts linked on each build:

• Classification and triage: 500 emails sorted for 3.5 cents (build); tickets, leads, and inboxes run the same pattern.
• Judging at scale: roughly 26,000 structured verdicts across 3,282 posts for $0.13 (build); the LLM-as-a-judge economics changed the week this shipped.
• Agent loops: a browser agent completing a flight search in about 7 seconds for $0.004, because choosing the next action stopped being the bottleneck (build).
• Real-time systems: sustained play of Doom at ~10 decisions per second (category), which is the latency claim tested somewhere it can't hide.

All numbers are as reported by their authors. The use-cases tour walks the full landscape.

What Jev is not

It does not write, summarize, translate, converse, or reason through open-ended problems. Pointed at any of those, it fails by design, and the failure is the spec working. Frontier chat models keep that half of the world; the strongest production systems use both, with Jev handling the high-volume easy decisions and a bigger model taking only the uncertain slice. The fraud-detection build is that architecture in one line: a reported 96/100 accuracy for about $0.07, because the expensive model never saw the easy cases.

There are also no official public benchmarks yet, and launch-period details (pricing, limits, access) live with TypeSafe at docs.typesafe.ai, which outranks anything written here. The honest limitations list is required reading before you build.

Where to go next

New here entirely: Start here routes you by goal. Ready to build: What is Jev Engineering is the skill, and the getting-started tutorial is the first afternoon. Evaluating: the review, the alternatives, and what builds actually cost.

Frequently asked questions

Is Jev an LLM?

Technically it's a language model, but built and served as a decision model: closed questions in, structured answers out, optimized for verdict speed and cost rather than text generation. Judge, not author. Under the hood, ecosystem clients describe a native Choice API: you supply the question and the candidate answers, and Jev returns a probability for each choice, which is where the confidence signals this site keeps citing come from.

Who makes Jev, and is this site affiliated?

TypeSafe AI makes Jev; official documentation is at docs.typesafe.ai. This directory is independent and not affiliated with TypeSafe: we catalog builds and keep the receipts.

What does Jev cost?

Officially, whatever TypeSafe's docs say today. In reported practice, real workloads have landed between $0.0004 and $0.13 for jobs spanning single drafts to tens of thousands of verdicts; the cost page keeps the ledger.

How fast is Jev?

Fast enough that builders run it inside game loops (a reported ~10 decisions per second in Doom) and live agent loops. Speed is the product; the real-time builds are the public proof.

Jev vs GPT: which do I need?

GPT-class models generate and reason; Jev decides. High-volume classification, routing, scoring, and judging favor Jev; writing and complex reasoning favor the big models; most serious systems cascade both.