What is Jev? How a Jev agent decides without writing a word
Jev is TypeSafe AI's decision model, the first "System One model": it answers typed questions about your text with a decision and a probability, never prose. See it answer below, then get an API key in a minute: no waitlist, no card.
Independent and unofficial. Jev is a model by TypeSafe AI; Jagent is not affiliated with or endorsed by them. Official documentation is at docs.typesafe.ai.
Score — a graded judgement
honestly whoever designed this checkout should be fired, worst thing I've used all year
Asking: How harmful is this content?
Level
Benign (0.33)
Confidence
0.67
Every answer here is a real Jev call. Run the same on your own data with a key: sign in and it is issued at once, no waitlist, no card. Get a key →
- 658
- developer accounts
- 84,243
- API calls served
- 162 ms
- median call, server side
- 13
- decision models, one key
Live from this site's own call log, refreshed hourly.
Use cases
What developers use Jev for
Twelve decisions Jev makes in real pipelines, each with the call we measured and a box that runs it again.
- MixedSupport ticket triageQueue, severity and churn risk from one ticket539 tokens · 515 ms
- ChoiceLLM model routingSend each prompt to the smallest model that can answer369 tokens · 458 ms
- ChoiceAgent tool selectionPick the agent's next tool; the LLM fills in arguments403 tokens · 507 ms
- MixedContent moderationYour rules, with a probability for each420 tokens · 531 ms
- MixedPhishing and email security classificationFive warning signs from one email, then a verdict533 tokens · 499 ms
- ScoreRAG passage classification and rerankingKeep passages that answer, drop the ones merely on topic364 tokens · 563 ms
- NoulAgent output guardrailsCheck a draft for leaks and promises before it is sent474 tokens · 499 ms
- MixedLead scoring and qualificationFit, buying intent and competitor mentions from a form472 tokens · 529 ms
- MixedInvoice classificationExpense category, a review flag and how soon to pay724 tokens · 632 ms
- MixedAI news filteringKeep or drop each headline for one reader's interests540 tokens · 595 ms
- MixedAI product categorizationCategory, use, audience and tags from one listing706 tokens · 605 ms
- MixedCSV data validationThe row checks a schema cannot express576 tokens · 634 ms
Skip Jev when you need prose, code or a conversation: it decides, an LLM writes.All use cases, side by side
Built with Jev
What developers built with Jev in its first week
Games it plays, browsers it drives, catalogues it scores: the authors' own videos.
How it works
What is Jev?
An LLM, what TypeSafe calls a System Two model, writes one token at a time and hands you a string: right for writing, slow and fragile when you only need to know which of five tools runs next.
Jev gives up writing entirely. You send one state plus any number of typed questions and get every answer back in one call, each with probabilities and a confidence; ten questions take about as long as one.
Versions, limits and price are on the Jev AI model card; request shapes on the Jev API reference.
Choice “Choose an option from a list.”
For routing, tool selection, intent classification, triage.
Returns choice · probabilities{option} · confidence
Score “Score the state on a rubric.”
For quality grading, ranking, rubric-based evaluation, content moderation tiers.
Returns score (0..n-1 weighted mean) · legend{index→label} · probabilities{index} · confidence
Noul “Is this statement true?”
For boolean gates, guardrails, verification checks, filters.
Returns noul (0-1 float) — and nothing else
Integrate
Call Jev from your code in one request
The same request in cURL, Python or TypeScript, with the answer it returned. Or keep the tools you already use.
curl -X POST https://jev-agent.com/api/v1/systemone \
-H "Authorization: Bearer $JEV_AGENT_KEY" \
-H "content-type: application/json" \
-d '{
"state": "Charged twice for September and cancelling Friday unless refunded.",
"questions": {
"route": {
"type": "choice",
"instructions": "Which team should handle this?",
"criteria": { "billing": "Payments and refunds", "technical": "Bugs and outages" }
},
"urgency": {
"type": "noul",
"instructions": "Does this need a reply today?",
"criteria": { "true": "Time-sensitive", "false": "Can wait" }
}
}
}'Response
{
"model": "jev-1.13.0",
"answers": {
"route": { "type": "choice", "choice": "billing", "confidence": 1,
"probabilities": { "billing": 1, "technical": 0 } },
"urgency": { "type": "noul", "noul": 0.96 }
},
"usage": { "input_tokens": 417, "…": "…" }
}Measured by other people
Independent benchmarks, not vendor multipliers
Three figures from studies this site did not run, each linked to its source and each with the caveat its author attached.
25×
faster
Found Jev roughly 25x faster and 580x cheaper than Claude Fable 5.1 — 0.35s versus 8.83s per passage.
Every · Speed/cost oriented; not an accuracy study.
580×
cheaper
Found Jev roughly 25x faster and 580x cheaper than Claude Fable 5.1 — 0.35s versus 8.83s per passage.
Every · Speed/cost oriented; not an accuracy study.
95.1%
accuracy, 5 signals combined
Jev's raw verdict loses clearly on accuracy against Claude Haiku 4.5 (McNemar p < 0.0001), while winning decisively on speed and cost.
anisselbd/jev-phishing-bench · Latency was wall-clock measured from a machine in France against US-hosted services.
Our own measurements, including the ones that went against Jev, are on the benchmarks page.
Built on Jev
Jev agent tools you can use today
A Jev agent asks Jev the questions and shows you the probabilities. These run in the browser with no code: paste a resume or an essay and read the verdict one criterion at a time.
- AI resume checkerA resume against a job description, scored on the questions a recruiter asks, each with its probability.
- Essay graderSix rubric criteria, a letter grade and the odds of a B or better, with the rubric in plain view.
- AI grader for teachersYour own rubric applied to a whole class set at once, exported as a spreadsheet.
- Rubric generatorA rubric from your assignment, then tested on one essay to flag the levels too close to call.
- SAT essay graderReading, Analysis and Writing each scored 1–4, as on the School Day essay, plus a total out of 12.
- Grading API and MCPThe same graders over HTTP, or as tools inside Claude, Cursor and other MCP clients.
Spec sheet
Jev at a glance
- Vendor
- TypeSafe AI
- Model class
- System One model
- Launched
- September 15, 2026 (out of stealth, $40M seed round led by DCVC)
- Model routes
jev-latestjev-previewjev-1.13.0- Input price
- $0.042 per 1M input tokens
- Output price
- Free (no output tokens are generated)
- Latency
- 70–500ms end-to-end
- Context
- 64k tokens
- Modalities
- Text only — no image or audio input at launch
- SDKs
- Python, JavaScript
- Open weights
- No — closed managed API only
FAQ
Frequently asked questions
What is a Jev agent?
Two things. Generically, an agent that hands its decisions to Jev (which tool runs next, where a request goes, whether an output passes a rubric) and leaves the writing to an LLM. It is also what people call this site, after its address: Jagent is an independent guide at jev-agent.com, not affiliated with TypeSafe AI, and its resume checker and graders are Jev agents you can use in a browser.
What is Jev?
Jev is TypeSafe AI's flagship model and the first of a class they call System One models. Instead of generating text token by token, it evaluates typed questions against a state and returns structured decisions — a chosen option, a score, or a probability — that your code can branch on directly.
Is Jev open source?
No. Jev is delivered exclusively as a closed, managed API with no open weights and no self-hosting or VPC option. Several community projects reimplement the interface with open models, but they are not Jev.
Can Jev hallucinate?
Jev does not generate free-form strings — it returns a choice, a score, or a probability from a set you define. Because the output space is constrained to options you supply, it cannot invent an answer outside that set. It can still be wrong or over-confident, which is why every answer ships with a confidence value.
How much does Jev cost?
$0.042 per 1M input tokens, and output is free because no output tokens are generated.
jev-agent.com is independent and unofficial. Authoritative docs live at docs.typesafe.ai.