What is Jev?

Jev is a model that answers questions, not prompts. You hand it a piece of state (a support ticket, a screenshot, a game frame, an order book) and a set of typed questions, and it hands back typed answers with calibrated probabilities. No prose, no parsing, no "as an AI".

Three primitives

Every question you can ask Jev is one of three shapes.

• Choice. Pick one option from a list. Returns choice, probabilities, confidence.
• Score. Rate the state on a rubric you define. Returns score, probabilities, confidence.
• Noul. Is this statement true? Returns a single noul value between 0 and 1.

You can mix all three in one call. A single request against a support ticket can ask which team it belongs to, how frustrated the writer is, and whether it is urgent, and get all three back together.

Why the catalog exists

The interesting number is never the benchmark. It is what a real build cost and how long it took. TypeSafe's own marketing quotes "193.6x faster, 444.6x cheaper" than a chat model on decision workloads. That is a claim; the catalog is the receipts. Every entry carries the cost and latency the author actually reported, or a dash where they did not.

What it is not

It is not a chat model and it does not generate text. If your job is to write, summarise or converse, Jev is the wrong tool. If your job is to decide (route, rank, filter, flag, classify, pick), it is the right one, and it is fast enough to sit inside a game loop or a browser agent's every step.

Where to go next

• Jev engineering: how the good builds structure their questions.
• Start here: the shortest path from zero to a working call.
• The catalog: what people have actually shipped.