Jev typed output
Possible outputs are defined before the call, so application code does not need to parse free-form prose.
JEV AI PLAYGROUND & JEV SCHEMA BUILDER
Build a Jev AI request in your browser
Explore the Jev playground for Choice, Score, and Noul—no account or API key required.
Unofficial community resource. Not affiliated with TypeSafe AI.
Local request builder
This builds a valid request locally; it does not call the Jev API or generate a model answer.
The shared text or structured context Jev evaluates.
Use letters, numbers, and underscores.
Ask one focused judgment about the state.
One option per line. Add an optional description after a colon.
Generated request JSON
{
"model": "jev-latest",
"state": "I was charged twice for order A-104. Please refund the duplicate payment today.",
"questions": {
"route": {
"type": "choice",
"instructions": "Which team should handle this request?",
"criteria": {
"billing": "Payments, invoices, or refunds",
"technical": "Bugs, outages, or integrations",
"sales": "Pricing, plans, or new accounts"
}
}
}
}Ready to run it? Use the official quick start
The playground runs entirely in your browser. Your state and questions are not uploaded.
JEV API PRICING
Jev cost calculator: compare input-token costs
Estimate monthly spend at current standard input rates for Jev, GPT-5.6 Terra, and Claude Sonnet 5.
Enter a token count or use K, M, and B shorthand.
tokens
Quick estimates
Estimated monthly input cost
At this volume
For 1M input tokens, Jev is 97.9% less and saves $1.96 versus GPT-5.6 Terra or Claude Sonnet 5.
Input-token prices only. Estimates use standard API rates and exclude LLM output tokens, caching, batch discounts, taxes, and provider fees.
Prices checked September 19, 2026.
WHAT IS JEV AI?
A decision model, not a chatbot
Jev is TypeSafe AI’s first System One model: it turns text or structured state into typed decisions and probabilities.
Instead of generating paragraphs one token at a time, the TypeSafe Jev model evaluates focused questions in parallel. That makes it useful for classification, routing, scoring, verification, and other decisions that software needs to consume directly.
Read the official System One overviewParallel questions
Multiple focused questions can share one state and be evaluated independently in the same request.
Calibrated uncertainty
Probabilities and confidence help software decide when to act and when to escalate for review.
$0.042 per million input tokens
TypeSafe currently charges for input tokens while Jev output tokens are free.
CHOICE · SCORE · NOUL
Jev’s three typed question primitives
Choose the primitive that matches the shape of the decision your software needs. Each returns a constrained result rather than generated text.
Choice
Which option fits?
Pick one item from a fixed set, such as routing a support ticket to billing, technical support, or sales.
Returns: choice, probabilities, confidence
Score
Where does it fall on a scale?
Evaluate ordered levels such as low, medium, high, and critical, with a probability distribution across the rubric.
Returns: score, legend, probabilities, confidence
Noul
How likely is this to be true?
Ask a clear yes-or-no question and receive the probability of yes, from 0 to 1.
Returns: noul probability
JEV API KEY & SDKs
Jev Python, Node, LangChain, and gateway quick starts
Get a Jev API key directly or use an established integration. These links cover the current Python example, Node SDK, LangChain package, OpenRouter route, and Vercel AI Gateway model.
Get a Jev API key
console.typesafe.ai
Sign in to the TypeSafe Console, create a direct API key, and use jev-latest with POST /v1/systemone.
Open TypeSafe quick startJev Python example
pip install typesafe-sdk
Use TypeSafeClient with Choice, Score, and Noul objects. The official package supports synchronous and asynchronous calls.
View the Python SDK exampleJev Node SDK
npm install @typesafe-ai/sdk
The official JavaScript and TypeScript client supports Node.js 20 or newer and infers answer types from your questions.
View the Node SDK exampleJev LangChain integration
pip install langchain-typesafe
LangChain exposes Jev through TypeSafeClassifier for typed classification inside agents, middleware, and routing flows.
Read the LangChain Jev guideJev OpenRouter access
typesafe/jev-1.13
Use an OpenRouter account and key to route requests to the current Jev 1.13 listing.
View the OpenRouter model pageVercel AI Gateway
typesafe-ai/jev
Use Vercel’s AI SDK evaluate API with a Vercel AI Gateway key and the TypeSafe provider route.
View the Vercel model pageCOPY-PASTE JEV CODE EXAMPLES
Run Jev with Python, Node, LangChain, or Vercel
These minimal examples follow the current provider documentation. Set the required API key in your environment before running them.
PythonJev Python example
pip install typesafe-sdkfrom typesafe_sdk import Choice, TypeSafeClient
with TypeSafeClient() as client:
response = client.system_one(
state="I was charged twice. Please refund the duplicate.",
questions={
"route": Choice(
instructions="Which team should handle this?",
criteria={"billing": None, "technical": None},
)
},
)
print(response.choices["route"].choice)Node.jsJev Node SDK example
npm install @typesafe-ai/sdkimport { choice, TypeSafeClient } from "@typesafe-ai/sdk";
const client = new TypeSafeClient();
const response = await client.systemOne({
state: { document: "I was charged twice. Please refund it." },
questions: {
route: choice("Which team should handle this?", {
billing: null,
technical: null,
}),
},
});
console.log(response.answers.route.choice);LangChainJev LangChain example
pip install langchain-typesafefrom langchain_typesafe import Noul, TypeSafeClassifier
classifier = TypeSafeClassifier()
response = classifier.invoke(
state="The deploy failed twice. Customers are seeing 500s.",
questions={
"urgent": Noul(
instructions="Does this need attention right now?"
)
},
)
print(response.nouls["urgent"].noul)VercelJev Vercel AI Gateway example
npm install aiimport { experimental_evaluate as evaluate } from "ai";
const result = await evaluate({
model: "typesafe-ai/jev",
state: "The support agent issued a full refund.",
questions: {
refunded: {
type: "boolean",
instructions: "Was a refund issued?",
},
},
});
console.log(result);JEV VS GPT-5.6 · JEV VS CLAUDE
Use Jev and LLMs for different jobs
The price gap is real, but this is not an equal-capability comparison. Jev specializes in fast, constrained decisions; general-purpose LLMs handle open-ended generation and reasoning.
| Dimension | Jev | GPT-5.6 Terra | Claude Sonnet 5 |
|---|---|---|---|
| Input price per 1M tokens | $0.042 | $2.00 | $2.00 |
| Output price per 1M tokens | Free | $12.00 | $10.00 |
| Primary output | Typed decisions and probabilities | Generated text and structured output | Generated text and structured output |
| Best fit | Routing, classification, scoring, verification | Reasoning, code, tools, broad generation | Reasoning, writing, code, agents |
| Not designed for | Chat, prose, code, long explanations | Ultra-cheap focused decisions | Ultra-cheap focused decisions |
Pricing uses current standard API rates. Real costs vary with output length, caching, batch processing, long-context tiers, gateways, and regional options.
INDEPENDENT JEV REVIEW
When Jev is the wrong tool
A constrained output can be structurally valid and still be the wrong judgment. Test accuracy and confidence thresholds on your own data before automating consequential decisions.
Review the official model limitsIt does not generate prose
Jev does not write replies, explanations, code, or creative content. Pair it with an LLM when you need generated language.
Typed does not mean factually correct
The answer stays inside your declared schema, but the selected option or probability can still be wrong.
English is currently strongest
TypeSafe says other languages, including CJK scripts, are supported but may not perform equally well.
High-stakes use needs safeguards
Use thresholds, deterministic checks, human review, and task-specific evaluation instead of treating confidence as a guarantee.
JEV AI FAQ
Frequently asked questions
Is this the official Jev AI website?
No. jev-ai.dev is an independent, unofficial calculator and guide. TypeSafe AI publishes Jev and maintains the official website and documentation.
What is Jev AI?
Jev is TypeSafe AI’s first System One model. It accepts text or structured state and returns typed Choice, Score, or Noul decisions with probabilities instead of free-form prose.
What is a System One model?
A System One model makes fast, focused judgments that software can use directly. It evaluates shared state against typed questions and returns constrained answers and probabilities rather than generated prose.
How much does Jev cost?
Jev 1.13 is currently priced at $0.042 per million input tokens. Output tokens are free. Gateway fees, taxes, discounts, and future pricing changes are not included.
Is Jev free?
Jev API usage is not generally free: TypeSafe lists a per-input-token price. This independent cost calculator is free and does not call the Jev API.
Can I try Jev free in this playground?
You can use this playground and schema builder free without an account. It creates valid Jev request JSON locally, but it does not run live Jev inference. A live call requires a provider account and API key.
How do I get a Jev API key?
Use the TypeSafe Console for a direct key, or create a key with Vercel AI Gateway or OpenRouter and use that provider’s Jev model ID.
Can I use Jev without the TypeSafe waitlist?
Jev is currently listed on Vercel AI Gateway and OpenRouter, which use their own accounts and API keys. Provider access and availability can change, so verify the current model page before integrating.
Does Jev hallucinate?
Jev cannot return a value outside the answer shape you define, but schema compliance is not the same as factual correctness. Its judgment can still be wrong, and calibration does not guarantee any individual answer.
What are Choice, Score, and Noul?
Choice selects one option from a fixed set. Score evaluates ordered levels. Noul returns the probability that a yes-or-no statement is true.
Can Jev replace GPT or Claude?
No. Jev is built for focused typed decisions. GPT and Claude remain better suited to chat, open-ended reasoning, writing, code generation, and explanations.
What are the main Jev alternatives?
Alternatives include structured-output calls to general LLMs, dedicated classification models, embeddings plus a classifier, and deterministic rules. The right choice depends on accuracy, latency, cost, and whether you need generated explanations.
What is Jev bad at?
Jev is a poor fit for free-form text, creative work, coding, long explanations, images, audio, video, and tasks that require slow multi-step reasoning inside one answer.
What are Jev rate limits and context limits?
TypeSafe currently documents 250,000 tokens per second, 1,200 requests per minute, a 64K total request budget, and a 32K limit for state plus the longest question. Rate limits may change without notice.
How accurate is this calculator?
It applies published standard input-token rates to the token volume you enter. It does not estimate output usage, tokenizer differences, caching, batch discounts, taxes, gateway fees, or model accuracy.
LAST UPDATED
September 19, 2026
Initial release with the local Jev request playground, browser-based cost calculator, copy-paste SDK examples, current Jev 1.13 pricing and limits, model comparison, limitations, and FAQ.
Want Jev updates and new tools? Join by email
Check the latest TypeSafe documentationSpot an outdated price or limit? Email a correction