Evaluation
LLM as a judge
Grade generated answers against your own rubric and get back a number you can sort.
- Score scales you define
- Probability on every tier
- Nothing to parse
LiveJev 1.13 and Jev latest answer here now
Jev AI Model Playground
Jev AI runs the Jev model, TypeSafe AI's System One decision model: ask typed questions about any text and get back probabilities, labels and scores. Try the Jev AI model free, then call the same API.
- 5 welcome credits
- 0.22 s median upstream answer
- Yes/no · Choice · Score
Inbound ticket
“We were charged twice for the September invoice and nobody has replied in four days. Refund the duplicate by Friday or we cancel.”
Jev 1.13 · real output · 1.8 s
- Choice100%
Which team should handle this ticket?
billing
- Yes / No
Is the customer asking for money back?
98% yes
- Score100% sure
How frustrated is the customer?
3.0 · Angry and ready to churn
Jev AI Use Cases You Can Run Now
Nine jobs teams hand to Jev AI instead of a general-purpose LLM. Each tab loads sample text and questions — edit either, run the Jev AI model, and read back answers you can drop into an if statement.
Decide which team owns an inbound ticket, how fast it has to be answered, and whether it can be closed by an automated reply — before a human reads it.
Model
1Text
Support ticket294 / 100,000
2Questions
4 in this requestEditing is free. Sign in and you come straight back here with your text and questions — no card needed.
3Answers
Sample · real Jev outputChoice100% confidence
Which team should handle this ticket?
billing
- billing100%
- technical0%
- sales0%
- account0%
- abuse0%
Score42% confidence
How urgent is this ticket?
1.58Today
Can wait for the normal queueWithin two business daysTodayBlocking revenue right now
Yes / No
Does this ticket need a human reply?
93%
YesYes / No
Is the customer asking for a refund?
99%
YesRecorded 2026-09-30 on typesafe/jev-1.13-20260917562 input tokensEdit the text or questions, then run for a live answer.
What Teams Build With the Jev AI API
Each guide covers the problem, the questions to ask Jev AI, a runnable request and the cost of one run.
Routing
LLM router
Pick the model, tool or queue for a request before a frontier model reads it.
- One label, every label scored
- 0.22 s median upstream
- Up to 20 questions a call
Retrieval
RAG evaluation
Check whether each answer is grounded in what you retrieved, cheaply enough to run on every query.
- Grounded / contradicted / unrelated
- Per-tier probabilities
- 32K context
What Is Jev AI?
Jev is TypeSafe AI's first System One model: send text plus typed questions, and every answer returns with a calibrated probability. Two Jev models answer here — Jev 1.13, a pinned build, and Jev latest, a rolling alias. jev-ai.org is a Jev AI playground and API.
Typed Answers, Not Text to Parse
Set the shape of every answer — a probability, a label, a point on your scale — and the Jev model returns exactly that, like a classifier you define in the request.
A Calibrated Probability on Every Answer
Each option comes back with its probability and a confidence, so you automate the clear cases and review the rest.
Twenty Questions in a Single Jev AI Request
Ask up to twenty typed questions about one input and pay a single input-token charge.
Fast Enough to Run Inline
A median Jev AI decision returns in about 0.22 s upstream — fast enough for routing and moderation queues.
Input Tokens Only, Output Free
The Jev model bills input tokens only, about $0.042 per million per TypeSafe, so high-volume scoring stays cheap.
One Request From Playground to Production
The body you build in the Jev AI playground is the one your code sends; the response shape matches upstream.
Three Answer Types in the Jev AI Model
You declare the shape in the request, and the Jev model cannot answer outside it.
Yes / no
"type": "noul"A probability between 0 and 1, not a bare boolean. Your code sets the threshold, so "probably" becomes a number you can tune.
Is this ticket urgent?
- noul
- 0.96
Choice
"type": "choice"One of the labels you declare, with a probability for every other label and a confidence for the pick. Nothing off-list comes back.
Which team owns this?
- billing
- 0.95
- technical
- 0.05
- sales
- 0.00
Score
"type": "score"A position on an ordered scale you define. Answers can land between tiers, so borderline cases stay visible.
How angry is the writer? (0–3)
- score
- 2.09
- confidence
- 0.88
One detail that costs people an afternoon: a choice takes its criteria as an object of label to description, while a score takes an array of tiers in order. Both are written out in the API documentation.
How to Use Jev AI in Four Steps
No waitlist and no SDK to install: the Jev AI playground and the Jev AI API share one endpoint and one request body.
- 01
Choose a Use Case
Open the playground and start from a scenario close to your problem. The sample text and questions are already filled in.
- 02
Edit the Questions
Paste your own text, then add, delete or retype questions. Twenty questions about one input still cost a single call.
- 03
Run the Jev AI Model
Sign in only when you want to run. Answers return in about a second, a run costs one credit, and a failed request costs nothing.
- 04
Ship It Through the API
Copy the matching request from the playground, create a key and send it to our endpoint — nothing to rewrite. Set up the Jev AI API.
Jev AI Plans and Token Prices
Pay for input tokens only; output tokens are free. Yearly plans cost half the monthly rate.
Price change coming
Current yearly rates end soon — get a full year of tokens at today's price.
Half price
Yearly plans · every token up front
Creator Yearly
$0.242 per 1M input tokens
$29$14.50
$174 charged once a year · $174 less than monthly
Works with
Playground, batch runs & API
Token allowance
720M input tokens for the year, all at once
Effective rate
$0.242 per 1M input tokens on the yearly plan
Billed on the input tokens each request actually uses. Output tokens cost nothing.
720M input tokens, delivered on day one
What's included
- Pay only for the input tokens you use
- Output tokens are free
- Unused tokens carry over
- Cancel before any renewal
MOST POPULAR
Studio Yearly
$0.148 per 1M input tokens
$49$24.50
$294 charged once a year · $294 less than monthly
Works with
Playground, batch runs & API
Token allowance
1.98B input tokens for the year, all at once
Effective rate
$0.148 per 1M input tokens on the yearly plan
Billed on the input tokens each request actually uses. Output tokens cost nothing.
1.98B input tokens, delivered on day one
What's included
- Pay only for the input tokens you use
- Output tokens are free
- Unused tokens carry over
- Cancel before any renewal
Max Yearly
$0.124 per 1M input tokens
$98$49
$588 charged once a year · $588 less than monthly
Multiplier
Works with
Playground, batch runs & API
Token allowance
4.752B input tokens for the year, all at once
Effective rate
$0.124 per 1M input tokens on the yearly plan
Billed on the input tokens each request actually uses. Output tokens cost nothing.
4.752B input tokens, delivered on day one
What's included
- 20% more tokens per dollar than Studio
- Pay only for the input tokens you use
- Output tokens are free
- Unused tokens carry over
Pay safely and securely with
Also available at checkout in supported regions: EPS, Kakao Pay, Naver Pay, PAYCO, Samsung Pay, MB WAY and SEPA Direct Debit. Checkout shows the methods that work for your country, currency and plan.
Common Questions About Jev AI
Quick answers about the Jev AI model, its API and TypeSafe AI.
What is Jev AI?
TypeSafe AI's first System One model. Instead of writing text, the Jev model answers typed questions — yes/no, a choice or a score — each with a probability.
What is Jev in AI, and is Jev an LLM?
People often call it the Jev AI LLM, but it never writes prose: every output is a typed decision, so there is nothing free-form to hallucinate.
Which Jev models can I use here?
Two. Jev 1.13 is a pinned build that will not change under you; Jev latest is a rolling alias that follows TypeSafe's newest release. Both take the same request in the playground and the Jev AI API.
How do I use Jev AI?
Pick a use case in the Jev AI playground above, edit the text and questions, and run it. Then send the same request through the Jev AI API.
Where can I try Jev AI?
Right here: sign in for five free credits and run the Jev AI model — no card, no waitlist. It is also on OpenRouter as typesafe/jev-1.13.
Does Jev AI have an API, and what does it cost?
Yes. The Jev AI API takes the same request as the playground. You pay for input tokens only; plans and token packs are listed below.
Can an AI agent use Jev AI as a tool?
Yes. Expose it as a Jev AI tool the agent calls before acting — to route a request, flag a risky action or check that an answer is grounded.
What is TypeSafe AI, and who is Diogo Almeida?
TypeSafe AI is the lab behind the Jev model. Founder Diogo Almeida worked at OpenAI on the research behind ChatGPT; Jev opened in early access on September 15, 2026.
Is jav ai the same as Jev AI?
Yes, jav ai is a misspelling of Jev AI. jev-ai.org is a Jev AI playground and API.
Try Jev AI Free in the Playground
Wondering how to try Jev AI? Sign in, take the five welcome credits and run it — no card required.