JEV

JEV (latest)

modeljev-latest

BudgetUnited States

Provider

JEV

Input $/1M

$0.427

Output $/1M

Free

About

JEV is a typed-decision API, not a chat model. You send a request shaped as {model, state, questions} — where state is the context to decide over and questions is the set of typed questions to answer — and get back structured JSON with a calibrated probability, category, or score per question, instead of free-form text you'd need to parse. It's built for pipelines that need a decision, not a conversation: ticket classification, routing, and scoring workflows where downstream code consumes the response directly.

Question types

What this model answers

noul

A yes/no binary question, answered with a calibrated probability rather than a flat label

choice

Picks one category out of a set of N you define (e.g. billing vs technical vs sales)

score

A numeric score on a scale you define (e.g. 1-5 frustration score)

Capabilities

What the API can do

●Answers noul (yes/no), choice (category), and score (numeric) questions in one call
●Returns a calibrated probability alongside binary answers, rather than just a label
●Accepts multiple questions per request against the same state
●Structured JSON output — no prompt engineering needed to parse the answer

Best for

Typical use cases

✓Customer-support ticket classification (routing by category)
✓Routing decisions — e.g. billing vs technical vs general inquiry
✓Sentiment or frustration scoring on incoming messages
✓Any workflow that needs a structured decision instead of free-form chat text

Limitations

Know before you integrate

!Typed-decision only — this is not a chat/completion model, and does not accept free-form conversational prompts
!US-hosted: the state you send egresses to the US for processing (mitigated by Anoman's PII masking applied to state before it leaves the region)
!Billed on input only — output tokens are free, so cost scales with the size of the state + questions you send, not the length of the answer

How it's measured

Calibration, not a leaderboard score

Each answer carries a calibrated probability or confidence value alongside its label, choice, or score — the model is designed to say how sure it is, not just what it picked. Anoman does not publish, and does not invent, accuracy or performance numbers for this model.

Compare across JEV models

Release channels, not performance tiers

jev-latest, jev-preview, and pinned releases like jev-1.13.0 are the same typed-decision API under different release channels — not separate products with different capabilities.

Model ID	Release channel	Input $/1M
jev-1.13.0	Pinned/stable release — deterministic behavior across upgrades of jev-latest / jev-preview	$0.427
jev-latest	Rolling alias — always tracks JEV's latest release; currently resolves to jev-1.13.0	$0.427
jev-preview	Early-access preview channel — currently resolves to jev-1.13.0	$0.427

Code example

Call jev-latest

curl https://api.anoman.io/anoman/v1/decisions \
  -H "Authorization: Bearer anm-sk-..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "jev-latest",
    "state": "Customer: My invoice charged me twice this month and I need this fixed today.",
    "questions": {
      "is_urgent": {
        "type": "noul",
        "instructions": "Is this ticket urgent and time-sensitive?"
      }
    }
  }'

See how Decision Models work Read the API docs