01Choice
Support ticket routing
Send every customer request to the team that can actually solve it.
Free Jev AI playground
Try Jev AI Free: The Jev Decision Playground
Jev AI is TypeSafe AI's System One Model. Instead of writing paragraphs, it answers a structured question with a typed decision and calibrated probabilities. Explore Choice, Noul and Score in a free Jev AI playground, no credit card needed.
Jev-AI.net is an independent Jev AI resource. It is not affiliated with, endorsed by or operated by TypeSafe AI. Jev is a product of TypeSafe AI.
statequestionoptionstyped decision + probabilities
01 / Jev AI playground
Jev AI Decision Playground
Describe the situation your software is looking at, ask one focused question and define the possible answers. Jev returns the answer it picks plus a probability for every option, so your code can act on it or hand it to a human.
Pick one answer from your options and get the full probability distribution.
Input / choice
01State
The context your software needs to evaluate.
02Question
03Choices
4 / 8A
B
C
D
Do not enter passwords, personal data or other sensitive information.
Output
ReadyDecision
—
Billing—
Technical Support—
Sales—
Account Security—
Ready when you are
Edit the input or load an example, then run a decision to see how Jev answers.
That is the idea behind Jev AI: software asks a bounded question and gets back a decision it can use directly, with a confidence it can check, instead of a paragraph it has to parse.
02 / Jev AI use cases
One pattern, many Jev AI decisions
Load a ready-made example into the playground to see how Jev handles routing, tool selection, risk checks and more.
02Choice
Agent tool selection
Let an AI agent pick the right tool for its next step.
03Score
Risk assessment
Score an action before it runs so risky steps get a human review.
04Choice
Agent loop control
Decide whether an agent should continue, retry, ask or stop.
05Noul
Urgency detection
Flag the messages that need a response right now.
06Score
Customer frustration score
Rate a message on an ordered rubric from calm to furious.
Question types
Three question types in Jev AI
Every Jev AI request pairs a state with a typed question. The question type decides the shape of the answer.
Choice
Pick exactly one option from a list you define. Jev returns the chosen option with the full probability distribution, and a single question can hold up to 255 options.
P(Billing) = 0.952
Noul
Ask a yes/no question about the state. Jev returns the probability that the answer is yes, a natural fit for guardrails, filters and smart if-statements.
P(yes) = 0.978
Score
Rate the state against an ordered rubric such as Low, Medium and High. Jev returns a probability for each level and a probability-weighted score you can threshold.
E[score] = 2.21 / 3
03 / A different kind of model
What is Jev AI?
Jev AI is the first public System One Model from TypeSafe AI, the lab founded by former OpenAI researcher Diogo Almeida, who helped build the methods behind ChatGPT. Jev was announced on September 15, 2026 and is rolling out through an early access program.
Where a large language model writes text one token at a time, Jev AI makes a fast, bounded judgment. You give it the relevant state and a structured question, and it answers with a typed value from the options you defined, plus calibrated probabilities that show how confident it is.
Because the answer space is fixed in advance, Jev cannot invent an option that does not exist and never returns malformed output. TypeSafe trained the model with a method it calls Reinforcement Learning for Calibrated Decisions (RLCD), so the probabilities are meant to be read as real confidence levels.
Traditional LLM
Text
Text
An open-ended response for a person to read.
Jev AI
State + question
Typed decision + probabilities
A bounded answer for software to act on.
04 / The decision path
How Jev AI works
Jev removes the steps between a question and an action. There is no text to parse and no schema to repair.
Traditional LLM workflow
6 steps- Input
- Prompt
- Generate text
- Parse
- Validate
- Decision
Jev AI workflow
3 steps- State + structured question
- Typed decision
- Probabilities
A simplified comparison. LLMs also offer structured outputs, and any production system still needs thresholds and a fallback for low-confidence answers.
05 / Jev AI at a glance
Why developers are excited about Jev AI
The numbers TypeSafe AI published at launch explain why Jev AI drew so much attention in its first week.
70–500 ms
70–500 ms responses
End-to-end response times of 70 to 500 milliseconds, which TypeSafe puts at 40x to 200x faster than frontier LLMs on equivalent decision tasks.
$0.042 / M
Input priced per billion tokens
Input costs $0.042 per million tokens ($42 per billion) and output is free, because a decision is only a few tokens long.
P ∈ [0, 1]
Calibrated probabilities
Every answer comes with calibrated probabilities and confidence scores, so you can set thresholds and send uncertain cases to a person.
0 type errors
No type errors
Outputs are typed values chosen from options you define in advance, so there is no malformed JSON and no hallucinated label to clean up.
≤ 255
Up to 255 options
A single Choice question can hold up to 255 options, enough for large routing tables, product catalogs or navigation tasks.
1 query
Parallel sampling
Jev uses a parallel sampler that produces all of its outputs in a single query instead of streaming tokens one by one.
Figures as published by TypeSafe AI at launch in September 2026. Real-world results depend on the task, and early access capacity is limited.
06 / Build with decisions
Jev AI API
Developers call Jev AI through TypeSafe AI's HTTP API and SDKs. A request carries the program state and one or more typed questions; the response carries typed values and probabilities your code can branch on directly.
Typical Jev jobs include ticket routing, spam and safety filters, agent tool selection, LLM output verification, lead scoring and map-reduce style classification over large datasets. To get access, join the early access program on the official TypeSafe AI website.
Decision pattern · illustrative pseudocode
state = "Refund request for a double charge"
question = Choice("Which team should handle this?",
["Billing", "Technical", "Sales"])
decision = jev.decide(state, question)
# -> Billing p = 0.95
if decision.probability > 0.8:
route_to(decision.value)
else:
send_to_human_review(state)Pseudocode for illustration only. See the official TypeSafe AI documentation for the real request format.
07 / Choose the right tool
Jev AI vs LLM
Jev is not a chatbot and does not replace a language model. It is a specialist for decisions, and the two work well together.
| Attribute | Jev AI | Traditional LLM |
|---|---|---|
| Primary job | Structured decisions | Text generation |
| Output | Typed decision + probabilities | Free-form text |
| Typical latency | 70–500 ms | Seconds, or minutes on multi-step tasks |
| Cost | $0.042 per million input tokens, free output | Priced per input and output token |
| Output reliability | Only options you define, no type errors | Can drift from the schema or invent labels |
| Best for | Routing, scoring, classification, guardrails | Writing, reasoning, coding, conversation |
A common pattern: let an LLM plan and write, and let Jev AI make the bounded calls inside the loop, such as which tool to use, whether an answer is safe and when to stop. Jev AI cannot explain its answers in words, so pair it with an LLM when you need a written rationale.
FAQ
Jev AI questions, answered
What you need to know about Jev AI, System One Models and this free Jev AI playground.
Try Jev AI FreeWhat is Jev AI?
Jev AI is TypeSafe AI's System One Model, a new kind of AI model that returns typed decisions with calibrated probabilities instead of generated text. It is built for fast, structured judgments inside software, such as routing, classification and scoring.
Who created Jev AI?
Jev AI was created by TypeSafe AI, a company founded by Diogo Almeida, a former OpenAI researcher who worked on the methods that made ChatGPT useful. Jev-AI.net is an independent site and is not affiliated with TypeSafe AI.
What is a System One Model?
System One Model is TypeSafe AI's name for models built for fast, intuitive, bounded judgments rather than slow, step-by-step reasoning. Jev AI is the first public System One Model.
Is Jev AI a large language model?
No. Jev is transformer-based, but it does not produce text. It reads text and structured state as input and outputs a probability for each allowed answer, so it cannot write essays, code or explanations.
What are Choice, Noul and Score?
They are the three question types in Jev. Choice picks one option from a list, Noul returns the probability that a yes/no question is true, and Score rates the state on an ordered rubric and returns a probability-weighted score.
What is Jev AI used for?
Common uses are customer support routing, agent tool selection, agent loop control, risk and safety checks, spam filters, lead scoring and verifying the output of other models. Real-time systems such as games, simulations and robots are a good fit too, thanks to the low latency.
How fast and how cheap is Jev AI?
TypeSafe AI reports end-to-end responses of 70 to 500 milliseconds, 40x to 200x faster than frontier LLMs on equivalent tasks. Input costs $0.042 per million tokens and output tokens are free.
Can Jev AI hallucinate?
Jev can only answer with options you define, so it cannot invent a label or return malformed output. It can still pick the wrong option, which is why the probabilities matter: send low-confidence answers to a person or a fallback.
Is this Jev AI playground free?
Yes. The Jev AI playground on Jev-AI.net is free and needs no credit card. The built-in examples work today, and live runs for your own inputs are being added.
How do I get access to the Jev AI API?
The Jev AI API is in early access. Join the waitlist on the official TypeSafe AI website; developers are being admitted in batches.
Try Jev AI free today
Load an example, change the state or the options, and see how a System One Model turns a question into a decision your software can use.