Decision Models

Decision models do not generate text. You give them a piece of content (the state) and a set of typed questions, and they return a typed answer for each question together with calibrated probabilities. They are built for the places inside an application where a fast, predictable, machine-readable verdict matters more than prose: routing, classification, triage, moderation gates, A/B triggers.

Compared with prompting an LLM for JSON, a decision model answers in a fraction of the time, cannot produce anything outside the question schema, and reports how confident it is instead of writing a number into free text.

Question Types

Type	Question	Answer
noul	A yes/no question.	noul — probability that the answer is "yes" (0–1).
choice	Pick exactly one option from criteria.	choice — the selected key, plus a probability for every option and a confidence.
score	Rate the state against ordered levels.	score — expected level index (may be fractional), a probability per level, the legend and a confidence.
box	Where something is in an image.	box — the four edges as fractions of the width and height, plus a confidence. Jev 1.14 only; experimental.

The state can be a plain string, a JSON object, or an array of texts. Images are supported by Jev 1.14 and d1, passed in assets rather than in state; audio and video are not supported.

The question types above are the ones Jev and d1 take. GPT-6 Luna answers on the same endpoint but with a different request shape: the content goes in input rather than state, questions is an array of named items rather than an object, and its yes/no type is predicate rather than noul. Sending one model the other's shape is a 400, so check the model's page before you switch model.

Pricing

Decision models are billed by input tokens only; output tokens are reported in usage but are free. See the pricing page and the model card for current rates.

Example: Triaging a Support Ticket

curl https://api.aimlapi.com/v1/decisions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <YOUR_AIMLAPI_KEY>" \
  -d '{
    "model": "typesafe/jev",
    "state": "Help! My payments have been failing for 3 days and nobody answers support.",
    "questions": {
      "is_urgent":   { "type": "noul",   "instructions": "Does this convey urgency?" },
      "department":  { "type": "choice", "instructions": "Which team should handle this?",
                       "criteria": { "billing": "Payments, invoicing, refunds", "technical": "Bugs, outages, integrations", "sales": "Pricing, upgrades, new accounts" } },
      "frustration": { "type": "score",  "instructions": "How frustrated is the customer?",
                       "criteria": ["Calm", "Frustrated", "Very angry"] }
    }
  }'

Every key from questions comes back under answers with a typed result:

{
  "model": "jev-1.14.0",
  "answers": {
    "is_urgent":   { "type": "noul",   "noul": 0.96 },
    "department":  { "type": "choice", "choice": "billing", "confidence": 0.98,
                     "probabilities": { "billing": 0.99, "technical": 0.01, "sales": 0 } },
    "frustration": { "type": "score",  "score": 1.21, "confidence": 0.69,
                     "legend": { "0": "Calm", "1": "Frustrated", "2": "Very angry" },
                     "probabilities": { "0": 0, "1": 0.79, "2": 0.21 } }
  },
  "usage": { "input_tokens": 406, "output_tokens": 73 }
}

All Available Decision Models

Model ID + API Reference link	Developer	Context	Model Card
typesafe/jev	TypeSafe AI	262 144	Floating — always the current Jev
typesafe/jev-1.14	TypeSafe AI	262 144	Current; adds images and box
typesafe/jev-1.13	TypeSafe AI	262 144	Previous release, pinned
liquid/d1	Liquid AI	32 768	No output tokens; reads images
openai/gpt-6-luna	Open AI	1 050 000	Different request shape; also a chat model