Jev Waiting List Removed: Start Free In 5 Mins

Q: Does Jev work with LangChain?

Yes. The langchain-typesafe package wires Jev in as a first-class component and reads your TYPESAFE_API_KEY from the environment. LangChain also shipped a ready-made model router built on Jev.

The Jev waiting list was removed on 21 September 2026, so you can grab $5 of free credit and have a working decision model running in about five minutes.

I am going to spend most of this post on what to do with those five minutes, because "the queue is gone" is a one-line story.

The interesting part is what a decision model priced at four cents per million tokens does to your AI bill.

I swapped one classification job off a frontier model and onto Jev, and the cost of that job dropped to something I stopped bothering to track.

That is the pitch, and the free credit means you can test the claim without entering a card.

Here is the quick start, then the three builds worth doing first.

The five-minute quick start

Go to console.typesafe.ai and create an account with your email.

Confirm the email and you will land inside the console, not on a holding page, because the queue is switched off.

Create an API key and copy it straight away, since the full string is only shown once.

Set it in your environment as TYPESAFE_API_KEY rather than pasting it into a file.

Install the official SDK for your language, or langchain-typesafe if you are already in a LangChain stack.

Send one Noul question — a plain yes or no against two lines of state — to confirm auth works before you build anything.

That is it.

If you registered during early access and never heard back, log in with the same email instead of making a second account.

Your $5 of credit should already be there.

What the $5 actually buys

Input costs $0.042 per million tokens, which is $42 per billion.

Output costs nothing at all, because the outputs are small enough that TypeSafe decided they were too cheap to meter.

So $5 gets you roughly 120 million input tokens.

For comparison, that same volume of input on a frontier text model would run into the hundreds of dollars.

That is not a trial — that is enough to run a real workload for a while and measure what it saves you.

What you pay	Jev	Typical frontier model
Input per million tokens	$0.042	Dollars, not cents
Output per million tokens	Free	Usually the expensive half
End-to-end latency	70ms–500ms	Often seconds
Free credit to start	$5 (~120M tokens)	Usually none

The full breakdown is in the Jev pricing post.

🔥 Want the exact setup I used to get these results? Inside the AI Profit Boardroom, I've got a full section on cutting AI costs — model routing, cheap classification layers and the stacks I run in my own business. Plus weekly coaching calls + 2,800+ members. → Get access here

Understand this before you spend a token

Jev is not a smaller ChatGPT, and people who joined the waiting list expecting one were disappointed.

Jev is a decision model.

You give it unstructured state plus a list of questions, and it returns typed answers with calibrated probabilities.

It never writes prose, and the output is designed for your code to consume rather than a human to read.

TypeSafe calls it a "System One" model — fast instinctive judgement rather than slow deliberate reasoning.

Three question types make up the entire API.

Choice picks one option from a list you supply, returning a probability for every option plus an overall confidence number.

Score rates the state against ordered levels you define, returning a continuous score plus a distribution across those levels.

Noul answers yes or no, returning a single probability that the answer is true.

Choice is capped at 255 options per query.

If you need to pick from thousands of items, score candidates in batches first and push the finalists through one Choice question.

I explained the model properly in the Jev AI model post.

Build 1: a model router that pays for itself

This is where I would spend the first hour.

LangChain shipped a ready-made router built on Jev, and the idea is embarrassingly simple.

You describe each of your models' strengths in plain English — this one is cheap and fast, this one is strong on code, this one handles huge context.

Jev routes every incoming request to whichever fits best.

Under the hood that is a Choice question answered in well under a second.

The saving comes from the requests that no longer hit your expensive model.

Most stacks send everything to the strongest model out of laziness, and most requests do not need it.

A router that costs four cents a million tokens to run sits in front of that and reroutes the easy ones.

Build 2: a triage layer for whatever queue you already have

If you run support tickets, leads, or an inbox, this is the second win.

Use Score rather than Choice, because severity is an ordered scale rather than a set of unrelated categories.

You define your levels, pass the message as state, and get back a continuous score plus a distribution.

The distribution is the bit people underuse — it tells you when the model is genuinely unsure rather than confidently wrong.

Route the confident ones automatically and send the uncertain ones to a human.

That one rule is the difference between an automation people trust and one they quietly turn off.

Build 3: a cheap safety gate on expensive actions

Put a Noul question in front of anything your agent does that costs money or cannot be undone.

"Is this action consistent with what the user asked for?"

You get a single calibrated probability back in under a tenth of a second.

Set a threshold, and below it you stop and ask.

At Jev's price this gate is effectively free, which is why it is worth adding everywhere rather than only on the scariest action.

More patterns are in the Jev use cases post.

🔥 Want to turn this into actual income? The AI Profit Boardroom is where I teach the full build — AI agents, automation and the offers that make them pay. Weekly coaching calls, 1,000+ AI agents, 2,800+ members. → Join the Boardroom

Why the waiting list existed at all

Worth a short detour, because the numbers are genuinely wild.

Jev launched on 15 September 2026, alongside a $40 million seed round led by DCVC.

Demand was heavy enough that TypeSafe briefly lost the ability to serve its own API during launch week.

Vercel had it live on AI Gateway a day later, and by hour 24 it was being used by nearly 13% of paid teams — more than twice the share of any previous model launch on the platform.

Vercel called it the fastest-adopted model in AI Gateway history.

Roughly 140,000 queued signups were processed inside about 36 hours.

Then on 21 September the queue came down entirely.

Six days from gated early access to open signup, driven purely by how many people wanted a decision model at that price.

The bits that are still unfinished

I would rather you heard this from me than found out in production.

TypeSafe's own disclosed figures put Jev at 67.8% aggregate accuracy against 74.1% for a comparable LLM on the same benchmarks.

The self-published JevBench score is 75.3, and self-published benchmarks deserve the obvious scepticism.

Rate limits are still being adjusted dynamically as GPU capacity arrives, with no published numbers.

TypeSafe has not said whether dropping the queue means general availability or an open beta.

There are no stated credit-expiry terms on the $5.

So wire in retries with backoff, keep a fallback path, and evaluate against your own data rather than the demo prompts.

Use it where a wrong call is cheap to correct, and keep a human on anything where it is not.

If you would rather not create a TypeSafe account

You do not have to, and you never did.

The same Jev 1.13 model runs on three gateways with no approval step.

Route	Model ID	What you need
OpenRouter	`typesafe/jev-1.13`	Prepaid OpenRouter credits
Vercel AI Gateway	`typesafe-ai/jev`	Card on the team
Cloudflare Workers AI	`typesafe/jev`	Cloudflare account

The answers are identical on all of them.

What you give up is the official SDKs, version pinning to a build like jev-1.13.0, status page access and the larger documented request budget.

For a prototype that is a fine trade.

For production I would go direct, especially now that there is nothing to wait for.

The detail is in the Jev AI API post.

Jev waiting list FAQ

Is the Jev waiting list gone for good?

TypeSafe removed the Jev waiting list on 21 September 2026 and signup has been open since. The company has not said whether this is full general availability or an open beta, so treat it as open but not formally guaranteed.

How fast can I actually get started with Jev?

About five minutes. Register at console.typesafe.ai, confirm your email, create an API key, set it as TYPESAFE_API_KEY, install the SDK and send one Noul question to confirm auth works.

What is the Jev free credit worth?

$5 per new account, which is roughly 120 million input tokens at $0.042 per million. Output tokens are free, so the credit goes a very long way on a decision model.

What should I build first with Jev?

A model router. Describe each of your models in plain English and let Jev pick per request — it is a single Choice question and it cuts the number of calls hitting your most expensive model.

Does Jev work with LangChain?

Yes. The langchain-typesafe package wires Jev in as a first-class component and reads your TYPESAFE_API_KEY from the environment. LangChain also shipped a ready-made model router built on Jev.

Can Jev handle more than 255 options in a Choice question?

Not in a single query. For larger option sets, score candidates in batches with Score questions first, then run the finalists through one Choice question.

About Julian

I'm Julian Goldie — AI entrepreneur, SEO expert, and founder of the AI Profit Boardroom (2,800+ members). I help business owners scale with AI agents, automation, and SEO.