Model reference · verified

What is Jev? TypeSafe's decision model, on OpenRouter

Jev is TypeSafe's first model, announced on 15 September 2026 and listed on OpenRouter three days later. It does not write. You hand it program state and a typed question, and it hands back a decision with a probability attached — the thing a chat model is worst at and a program needs most. This page carries the specs, the price, the three question types, and a clear list of what it is not.

Reference Listed 2026-09-18T00:01:24Z Verified 2026-09-18 OpenRouter listing →

The short version

Maker

TypeSafe AI

Announced 2026-09-15

Class

System One

Decisions, not text

OpenRouter ID

typesafe/jev-1.13

Listed 18 Sep 2026

Context

32,000

tokens

Input price

$0.042

per million tokens

Output price

Free

returns a decision

Latency

70–500 ms

vendor-claimed

Writes text?

gives up string generation

Verified specs

Read on 2026-09-18 from the OpenRouter endpoint, the Cloudflare Workers AI model card and TypeSafe's own announcement. Every row can be re-checked at the links below; the vendor's claims are labelled as the vendor's claims, including the ones we would like to be true.

Field	Value
OpenRouter ID	typesafe/jev-1.13 Display name in the picker is “TypeSafe: Jev 1.13”.
Maker	TypeSafe AI Announced by the founder, Diogo Almeida, on the TypeSafe blog on 2026-09-15.
Class	System One model TypeSafe’s own category: models built to return decisions rather than prose.
Input / output	Text (program state) in → typed decisions out The catalogue field reads `text->decisions`. It is not a chat endpoint.
Context window	32,000 tokens Read from the OpenRouter endpoint and the Cloudflare Workers AI model card; both agree.
Price — input	$0.042 per million tokens $0.000000042 per token. The vendor quotes the same figure as $42 per billion tokens.
Price — output	Free The catalogue lists completion price 0, and the vendor says "output tokens: FREE (too cheap to meter)".
Serving provider	TypeSafe Single endpoint, provider tag TypeSafe — the maker serves it directly.
Question types	Noul, Choice, Score Noul is the yes/no form (criteria you name, answered as a probability); Choice picks from a named set; Score grades on an ordered scale.
Every answer carries	A probability and a confidence Cloudflare’s model card: “returns calibrated answers with probabilities and confidence”.
Training method	RLCD — Reinforcement Learning for Calibrated Decisions The vendor’s name for the method; the alternative they describe it against is RLHF/RLVR.
Vendor speed claim	70 ms – 500 ms end to end Against “3 to 329 seconds for frontier models”, which the vendor frames as 40×–200× faster on System One shaped queries. Their own number, not an independent benchmark.
Availability	Early access — waitlist The 2026-09-15 announcement calls Jev “available today in early access”; TypeSafe’s site carries a Join Waitlist link.
Also listed on	Cloudflare Workers AI Model id `typesafe/jev`, described there as a third-party model. https://developers.cloudflare.com/ai/models/typesafe/jev/

Source openrouter.ai/api/v1/models/typesafe/jev-1.13/endpoints Verified 2026-09-18 Announcement typesafe.ai

What it actually returns

Every call is a state plus a set of named questions. There are three question types, and the shape of each answer is fixed in advance — which is what makes the output safe to hand straight to code instead of parsing it and hoping.

Type	Comes back as	Use it for
Noul	a probability between 0 and 1	You name the true and false criteria; the model answers how likely the true one is.
Choice	one option from a set you define	Routing, classification, triage — wherever the branches are already written down.
Score	a position on an ordered scale you define	Grading, priors, severity. The scale is yours, so the output is already meaningful to code.

Source developers.cloudflare.com/ai/models/typesafe/jev Vendor framing the launch post

Figure 1 — why this is not a chat call. A chat model returns text that code has to parse and validate; Jev returns the decision, already typed, with the uncertainty attached. Diagram ours; the latency band is the vendor's claim.

Figure 2 — the gap is real, and it is also the maker's own measurement, run from their machines on System One shaped queries. We have not reproduced it. Note the axis: at this scale the difference looks like roughly a decade, which is the same thing the vendor calls 40×–200×.

Why this model is being searched for

The bare word “jev” sat flat for three months and then moved within a day of the announcement. The interesting part is not the size of the spike — it is which words come with it.

2026-09-15

TypeSafe announces System One models and Jev

Diogo Almeida, founder, publishes the launch post: two years in stealth, a new architecture with a parallel sampler, and RLCD as the training method. Jev ships in early access with a waitlist.
2026-09-15 → 16

Search interest leaves the floor

Worldwide Google Trends index for “jev” had sat flat for three months. It moves 1 → 4 on 15 September and 4 → 26 on 16 September — the first time the term has moved at all in the window.
2026-09-17

Interest doubles again, and the AI reading wins

Index 26 → 61. The related queries that appear are all AI-side: “ai jev”, “jev typesafe”, “jev model ai”, “typesafe ai”, “llm jev”, “diogo almeida”.
2026-09-18 00:01 UTC

Jev is listed on OpenRouter

The catalogue returns `typesafe/jev-1.13` with a creation time of 2026-09-18T00:01:24Z, 32,000 context and input at $0.042 per million tokens with free output. “openrouter jev” becomes a breakout rising query the same day.
2026-09-18

The term peaks, and the reference pages land

Index reaches 100 — the high point of the trailing three months, up from a floor of 1. Cloudflare’s Workers AI model card and a LangChain harness guide both go up the same day.

Source Google Trends, worldwide, trailing 3 months Read 2026-09-18 Index is relative to the term's own peak, not a search volume

What the community claims

None of this comes from TypeSafe, OpenRouter or Cloudflare. It is here because it is what people are actually saying, with the evidence attached so you can weigh it — and because the loudest claim of the week is the one nobody can check.

“Up to 190× faster and 440× cheaper than an LLM.” Medium confidence

This circulates on X as a summary of the vendor’s own figures. The arithmetic follows from the published ranges (70 ms vs 3 s, $0.042 in / $0 out vs $0.20–$10 in), but the baseline is the vendor’s, chosen to compare a decision call against a chat call. Treat it as the vendor’s comparison restated, not as an independent benchmark.

People are using Jev to drive website operations and automated trading. Low confidence

Reported in the X timeline on 2026-09-18, in threads that are hours old. We found no published workflow, repo or result that we could check, and no returns claim can be verified at all. Directional chatter about a three-day-old model, nothing more.

Jev is the start of a split in how AI gets used, not another chatbot. Low confidence

The most-shared take on X and LinkedIn on 18 September, arguing that decision models will absorb the automation work LLMs were being asked to do. It is an argument, not a measurement — but it is the reason the term is being searched, so it is worth knowing it exists.

How to call it

Get access. Jev ships in early access, and TypeSafe's site runs a waitlist. OpenRouter and Cloudflare both list it already, so an account on either side is the shorter path.
Pick a front door. Same model id on both; Cloudflare is the shorter hop if your code is already a Worker.
Send state, not a prompt. The input is program state — the thing your code already has — plus questions with the criteria named in them.
Read the probability, not just the label. Every answer carries a confidence. Threshold it in your own code; that is the whole point of the design.

Where	Model id	Note
OpenRouter	`typesafe/jev-1.13`	One key, one bill, the same endpoint the rest of the directory is read from.
Cloudflare Workers AI	`typesafe/jev`	Listed as a third-party model; called through env.AI.run with a state and a questions object.

How it differs from the models in our directory

	Chat LLM	Jev
Sends back	Text you parse and validate	A typed value, already valid
Says when it is unsure	Only if prompted, and inconsistently	Every answer carries a probability
Output cost	Roughly 5× the input price	Free
Latency	Seconds — 3 to 329 s per the vendor	70–500 ms, vendor-claimed
Can it explain itself?	Yes	No — and that is the trade

Comparison built from TypeSafe's own table Vendor claims are marked as such

Four other things called Jev

Three letters, four unrelated meanings, and all of them outrank this model on search volume. If you searched “jev” and wanted one of these, the table is the fastest way out — and if you wanted the model, it is the reason to add a qualifier.

Also “jev”	What it is
jev.	Jephté Kewto, a Congolese-Canadian rapper — Wikipedia entry, ~1.6M monthly listeners on Spotify. The largest legitimate non-AI meaning.
FaZe Jev	A YouTube channel with millions of followers. “faze jev” is the second-largest related query for the term.
JEV	Japanese encephalitis virus. “jev vaccine”, “jev symptoms” and “jev virus” are the medical cluster, served by WHO/CDC-grade pages.
jev (Czech/Slovak)	An ordinary word meaning “phenomenon” — “meteorologický jev”, meteorological phenomenon, is a standing Czech search. A reminder that the three-letter string has a life of its own outside English.

What we checked ourselves

This page is not a summarised press release. Below is every read we made on 2026-09-18, what we asked, and what came back — including the one thing we could not check.

Question	How we asked	What came back
Is Jev really on OpenRouter?	`GET /api/v1/models/typesafe/jev-1.13/endpoints`	200. Provider TypeSafe, modality `text->decisions`.
When was it listed, and where does that timestamp come from?	the `created` field of the same response	`1789689684` — 2026-09-18T00:01:24Z. Read from the catalogue, not from the announcement.
What does it actually cost?	the `pricing` block of the same response	`prompt 0.000000042` per token (= $0.042 per million), `completion 0`.
Do the two catalogues agree on context?	OpenRouter endpoint vs the Cloudflare model card	Both say 32,000. When two independent listings agree, we print the number without a hedge.
Why is this page hand-written?	grep the whole `/api/v1/models` list (445 entries) for `jev` and `typesafe`	No match. It is served, and absent from the list our sync reads. That is the finding this page exists to record.
What are the question types, exactly?	the code sample on the Cloudflare Workers AI model card	`noul` with named true/false criteria, `choice` with named options, `score` with an ordered list.
Is the speed and price claim real?	the vendor's launch post, 2026-09-15	They publish 70–500 ms, $0.042 per million in, output free. We have not reproduced any of it — so it is labelled as theirs everywhere on this page.
Is anyone actually searching for it?	Google Trends, worldwide, three months, daily	Index 1 for 92 of 93 days, then 4 → 26 → 61 → 100. Related queries are AI-side, not the rapper.
Does it work?	a live call	Not checked. We have no key for the early-access model, so nothing on this page comes from a real run — no sample output, no measured latency, no probability we observed ourselves. When we get access, this row becomes the first figure on the page.

Read 2026-09-18 Requests openrouter.ai/api/v1 These are our reads, not the vendor's claims

FAQ

What is Jev?

Jev is a model from TypeSafe AI, announced on 2026-09-15. Instead of generating text it takes program state in and returns typed decisions out — a yes/no with a probability, a choice from a set you define, or a score on a scale you define. TypeSafe calls this class a System One model.

Is Jev on OpenRouter?

Yes. The catalogue serves `typesafe/jev-1.13` — 32,000 context, $0.042 per million input tokens, output free — and the model page is at https://openrouter.ai/typesafe/jev-1.13. One caveat worth knowing: it answers on its own endpoint but is not returned by the `/api/v1/models` list, so tools that build off that list will not see it.

Is Jev an LLM? Can it write text or code?

No. It gives up string generation entirely — that is the trade the vendor made for speed and for never producing a type error. If you need prose, code or a chat answer, Jev is the wrong tool; it is built to be called by code that already knows what it is going to do with the answer.

What is a System One model?

TypeSafe’s term for a model trained to make fast, structured decisions rather than to produce human-pleasing text. The vendor’s framing is a function call with frontier intelligence behind it: unstructured state in, typed probabilistic decisions out. The training method is named RLCD — Reinforcement Learning for Calibrated Decisions.

What do Noul, Choice and Score mean?

They are the three question types. Noul is the boolean form — you name the true and false criteria and get back a probability between 0 and 1. Choice asks the model to pick one option from a set you define. Score asks it to place the answer on an ordered scale you define. Every answer comes with a probability and a confidence.

How much does Jev cost?

$0.042 per million input tokens on OpenRouter — the vendor quotes the same as $42 per billion — and output is free. Read the input figure as the whole bill: the model returns a decision, not a body of text, so there is no large completion to pay for.

How fast is Jev?

The vendor claims 70 ms to 500 ms end to end, against 3 to 329 seconds for frontier LLMs, and describes that as 40× to 200× faster on System One shaped queries. That is the maker’s measurement, run from their own machines; we have not reproduced it.

Is Jev free?

Output is free and input is $0.042 per million tokens, which is close to free in practice — a million decisions costs about four cents of input. It is not a free-preview model though: this is the listed price, not a temporary window like the Alpha-line stealth releases had.

Can I run Jev on Cloudflare Workers AI?

Yes — it is listed in the Cloudflare Workers AI catalogue as the third-party model `typesafe/jev`, with a code sample that calls it through `env.AI.run` and passes state plus a questions object. Same model, different front door from OpenRouter.

Is this the same Jev as FaZe Jev, the rapper jev., or the JEV virus?

No. All four share three letters and nothing else. Those other meanings have far more search volume than the model does — which is exactly why this page names them instead of quietly counting their traffic as ours.

Can Jev be used for automated trading?

It is being discussed that way on X, and a model that returns calibrated probabilities in under a second is a plausible fit for decision points in a trading loop. But we could not find a single published, checkable result — and nobody publishes a verified returns claim. Treat it as a rumour about a three-day-old model. This page will say so until there is something to link to.

When the next model lands, you will hear about it

Jev arrived with an announcement, a company and a founder — unlike the Alpha line, which shows up anonymous and free. We track both. Leave an email and we will send one note when the next one appears, with the specs read from the catalogue.

One email per model. No newsletter, no forwarding, unsubscribe in one click.