Looking for more models?Explore decisions-api.dev
Jev AI Model · STRUCTURED DECISIONS · 100 WELCOME CREDITS
Classify, score, and route text with Jev AI Model. Start from a real task, inspect the probabilities, and compare your rules before connecting the API. New accounts receive 100 credits.
ILLUSTRATIVE EXAMPLE · NO CREDITS USED
“I was charged twice this morning. Please refund the duplicate payment today.”
Choice
Billing
Route to the billing queue
Yes / no
96%
Probability of an urgent request
Score
2.8 / 3
Urgency on a 0–3 scale
Static illustration of the output format. Values are not a live model response or a measure of accuracy.
01 / WORKBENCH
Load a task without signing in. Edit the input, then sign in to run it with credits. New accounts receive 100 credits once.
What do you want to decide?
Choose a task to load its input and rules. Loading examples uses no credits.
We reserve this budget before the model runs, then charge actual usage and return the difference. Failed calls are refunded.
New account? Receive 100 credits once. Short requests typically use 1 credit; longer inputs can use more.
Draft saved in this browser · Your draft will be restored after sign-in.
ILLUSTRATIVE EXAMPLE · NO CREDITS USED
“I was charged twice this morning. Please refund the duplicate payment today.”
Choice
Billing
Route to the billing queue
Yes / no
96%
Probability of an urgent request
Score
2.8 / 3
Urgency on a 0–3 scale
Static illustration of the output format. Values are not a live model response or a measure of accuracy.
02 / COMPARE BEFORE YOU SHIP
One convincing answer is a useful beginning, but it does not tell you how a decision behaves across your traffic. Develop a Jev AI Model workflow with representative inputs, explicit expected outcomes, and a repeatable comparison. The workbench combines editable questions, saved configurations, run history, and a small evaluation workspace to support that process.
Open the full workbenchStart with the action your application needs to take. In Jev AI Model, name the question clearly, choose its type, and describe the criteria in language another person could apply. Add examples that sit outside the normal case. If a ticket mentions both billing and login, specify which issue should control its route.
Inspect the answer, available probabilities, and request details together. When a Jev AI Model result looks surprising, first check whether the input contains the evidence your criteria require. A high probability is not measured accuracy. It needs to be checked against labeled examples from your own domain before you use it as an automation threshold.
Capture configuration A, revise the question, then capture configuration B. The evaluation workspace in Jev AI Model runs both against the same supplied states, with up to 25 cases in a comparison. Add expected answers where available, inspect individual failures, and export the report. A change that fixes one ticket may still make another class of tickets worse.
Save the configuration you want to keep, revisit private web run history, and rerun difficult cases after an edit. Before connecting Jev AI Model to a live workflow, decide how your application handles missing answers, uncertain results, and request errors. Repeat the evaluation when your policy, input format, or model choice changes.
Evaluation separates model behavior from application policy. Jev AI Model produces the decision signal; your product decides what happens next. Set a review path for cases that should not proceed automatically, and judge the workflow by the outcomes that matter to your users.
03 / START FROM A REAL TASK
These editable recipes show how to connect a question to an application decision. Open a recipe in Jev AI Model, replace its sample state with your own text, and adjust the criteria to match your product. Each example is a starting point for testing, not a ready-made policy for every business.
A customer reports a duplicate charge. Use Jev AI Model to distinguish a billing issue from a technical fault or an account access problem, then pass the selected label to your support system. Include enough context to separate a refund request from a failed checkout.
Try ambiguous tickets as well as clear ones. Keep an other category for requests outside the available teams, and decide when uncertain results should reach a person.
Open and customize recipeSome requests need a record lookup; others need a language model or human review. Jev AI Model can classify the incoming task against routes you define, giving your application a structured signal before it starts the next operation.
Describe each route's capabilities and limits. Your application remains responsible for executing tools, checking permissions, handling failures, and deciding whether the proposed route is appropriate.
Open and customize recipePlace a claim beside the source passage and ask whether that passage supports it. With Jev AI Model, a citation check becomes a specific decision whose input and result can be reviewed together, rather than a vague request to judge an entire document.
Test missing evidence, partial support, and direct contradictions. The supplied passage sets the scope: this check does not independently browse the web or establish that the source itself is correct.
Open and customize recipe04 / UNDERSTAND THE OUTPUT
Jev AI Model gives developers a place to turn text and context into decisions that software can use. Describe the situation, define the question, and inspect the returned answer and probabilities. A support ticket becomes a route; a policy check becomes a yes-or-no judgment; a quality review becomes a score with explicit criteria.
The underlying Jev model is a System One model from TypeSafe AI. This independent service brings that decision workflow into a browser workbench and a paid API. With Jev AI Model, you can develop the question interactively, keep a useful configuration, and carry the same request structure into your application once you have tested it.
Use a choice question when your application needs one label from a defined set. In Jev AI Model, each option has a name and a description, so you can explain what belongs in billing, technical support, or account access. Inspect the probability distribution as well as the selected label, especially when two options overlap.
Write distinct criteria and include an explicit fallback for inputs that do not fit. The result can inform a queue assignment, a search filter, or the next step in an agent.
A score question is useful when a label alone loses information. Define ordered levels for relevance, urgency, or completeness, and explain what each level means. Jev AI Model returns a score along that scale, making it possible to compare inputs using a shared rubric instead of an open-ended written opinion.
Keep the order consistent and test examples near each boundary. A score only becomes useful when the people and systems consuming it agree on what the scale represents.
Use noul for a focused yes-or-no question, such as whether a source supports a claim or whether a request needs review. The yes probability from Jev AI Model lets your application choose a threshold. It expresses the model's judgment about the supplied evidence; it is not a guarantee that the statement is true.
Ask one question at a time and describe how missing evidence should be treated. Keep consequential actions behind your own business rules and review process.
Better inputs begin with the decision you need to make. Give Jev AI Model the relevant message, policy, and known facts, while leaving out unrelated conversation. Text, JSON objects, and arrays can organize that context. The workbench helps you inspect the request before running it, so you can see exactly what the model will receive.
Designed for developers
Move a tested question from the browser into your backend using the same state and question structure. The Jev AI Model workbench provides request examples in cURL, JavaScript, and Python, so you can inspect the payload before connecting it to an existing service. Create an API key in your account and send it as a Bearer token from your server.
A successful Jev AI Model API response includes structured answers and request usage information. Check the response status before reading the answers, record the request identifier for troubleshooting, and handle errors explicitly. Keep the API key out of browser code and public repositories. Your integration controls the next action, whether that is assigning a queue, requesting a review, or invoking another tool.
Web, API and evaluation share one credit balance and the same per-model rates. We reserve a budget before each call and settle actual input usage on success. Failed calls are refunded. Inspect the generated code and request examples before integrating.
POST /v1/systemone
typesafe/jev-1.13 · liquid/d1 · upstage/solar-decide · jaredpalmer/kev-4b · togethercomputer/tev1-4b-experimental · laya-english · laya-multilingual · laya-auto
// Server-side JavaScript (Node 20+ or Bun). Set JEV_API_KEY.
const response = await fetch("https://jevaimodel.net/v1/systemone", {
method: "POST",
headers: {
"Authorization": `Bearer ${process.env.JEV_API_KEY}`,
"Content-Type": "application/json",
},
body: JSON.stringify({
"model": "typesafe/jev-1.13",
"state": "I was charged twice for order A-4471. Please refund the duplicate payment.",
"questions": {
"route": {
"type": "choice",
"instructions": "Which team should handle this ticket? Use other when no option fits.",
"criteria": {
"billing": "Payments, charges and refunds",
"technical": "Bugs and product errors",
"account": "Login and account access",
"other": "None of these teams"
}
}
}
}),
});
const body = await response.json();
if (!response.ok || body.code !== 0) throw new Error(body.message);
console.log(body.data.result.answers);
console.log(body.data.creditsUsed);05 / UNDERSTAND THE SERVICE
Create an account to receive 100 credits once. Load a recipe, review the budget and run your questions in Jev AI Model. Short requests typically cost 1 credit; longer inputs can cost more. Saved configurations and private history help you build on earlier work.
Buy credits when you need more. Every pack works across the Jev AI Model playground, API and comparison tool at $1 per 10,000 credits. Credits do not expire. Successful requests are billed on input usage, with a minimum of 1 credit per request.
The Jev AI Model account area separates web activity from API usage and gives you a credit history to review. A request identifier helps connect an API call to its usage record. Private web history stores the state and answers so you can revisit a decision; API usage records track request metadata without storing the full request and answer bodies.
One credit balance for web, API and evaluation. New accounts receive 100 credits.
$1 = 10,000 credits on every pack. Credits do not expire. Pay once, with no subscription. Successful calls start at 1 credit; actual input usage determines the charge. Site input rates per million tokens: Jev / Kev $0.06, Solar Decide $0.072, Span-01 $0.03, and Laya $0.06. Liquid d1 currently has no token-based charge and uses the 1-credit minimum per successful call. Output tokens are not charged.
Starter
For your first workflows and integration.
Pro
For regular use and repeated evaluations.
Volume
For higher usage at the same unit rate.
Estimate with your average input tokens, including state and questions. Each request is rounded up separately.
Credits per request
1
Estimated total credits
10,000
Credit value at selected pack rate
$1.00
max(1, ceil(input_tokens × 600 / 1,000,000))
All live calls use credits, including web and evaluation. Minimum 1 credit per successful request; output tokens are not charged. Span-01: 300 credits/million input tokens. Span-01 Lite and currently free Liquid d1: 1 credit per successful request. Jev, Kev, Tev and Laya: 600 credits/million; Solar Decide: 720 credits/million. $1 buys 10,000 credits. Failed calls are refunded. Estimates exclude taxes.
Jev models benchmark
Use independent benchmarks as background when evaluating Jev AI Model. The linked comparison identifies its source, model versions, and snapshot date, so you can understand what the published scores describe. A benchmark result is evidence about that evaluation, not a promised accuracy rate for your own tickets, policies, or documents.
For a deployment decision, pair those results with your own labeled cases in Jev AI Model. Compare answer quality, uncertainty, response time, and observed credit use on inputs that resemble the work your product will actually receive.
Jev 1.13.0
#1 · JevBench v1.3.0
74.4
Score
SemIf
#2 · open rebuild
73.1
Score
djev
#3 · hosted preview
73.0
Score
JevBench v1.3.0 · 52 systems · 534 decisions · snapshot: Sep 21, 2026
Find the right starting point for Jev AI Model, whether you are exploring Jev AI for the first time or adding a Jev model decision to an existing application.