Explainer
What are decision models?
A decision model is a model that answers typed questions about a piece of text — a yes/no question, a choice between options or a score — and returns probabilities instead of free text. Decision Models hosts such models behind a Jev-compatible API.
What a decision model is
You send a state (the text or data to judge) and one or more typed questions. The question types are noul (yes/no), choice (pick one of your options) and score (rate against your own scale). The answer is typed too: for a choice it holds the chosen option, a confidence and a probability for every option. Output tokens are free because the API returns decisions rather than generated prose.
How it differs from a chat LLM
A chat LLM writes text, and you have to parse that text to get a label. A decision model returns the label and its probabilities in a fixed structure that your code can read directly, and your code can act on the probabilities, for example by sending uncertain cases to a person. It does not produce free-form answers or explanations.
When to use one
Decision models fit decisions that a program has to make many times: routing a request to the right queue or model, moderating content against your own policy, classifying text and scoring a draft against a rubric. See the use cases for LLM routing, content moderation and support ticket routing.
How decision models are measured
JevBench is a public benchmark of decision models on Benchmark Heaven; its capability score averages intelligence and calibration. See the current results on benchmarkheaven.com/jev-models. For the models we serve, see our benchmark sources and the models page.
A minimal API example
This is the request shown on the mammoth-1 model page. Replace $DECISIONMODELS_API_KEY with a key from your dashboard.
curl https://api.decisionmodels.io/v1/systemone \
-H "Authorization: Bearer $DECISIONMODELS_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model":"mammoth-1","state":"Mia owns a red bicycle.","questions":{"color":{"type":"choice","instructions":"Which color is the bicycle?","criteria":{"red":null,"blue":null}}}}'{
"id": "dec_example",
"model": "s1-fast",
"answers": {
"color": {
"type": "choice",
"choice": "red",
"confidence": 0.91,
"probabilities": { "red": 0.91, "blue": 0.09 }
}
},
"usage": { "input_tokens": 328, "output_tokens": 0, "decisions": 1 },
"tier": "eu"
}Where to go next
Frequently asked questions
What is a decision model?
A decision model answers typed questions about a text — yes/no, a choice between options or a score — and returns probabilities instead of free text.
How is a decision model different from a chat LLM?
A chat LLM generates text that you have to parse. A decision model returns a typed answer with probabilities for each option, and it does not generate free-form answers or explanations.
What are decision models used for?
Routing requests, moderating content, classifying text and scoring drafts against a rubric.
What is JevBench?
JevBench is a public benchmark of decision models on Benchmark Heaven. Its capability score averages intelligence and calibration.
Is there a Jev-compatible API for decision models?
Yes. Decision Models is independent and offers a Jev-compatible API: you send a state and typed questions, and you get typed answers with probabilities. Jev is a trademark of TypeSafe AI; Decision Models is not affiliated with TypeSafe AI.
Jev is a trademark of TypeSafe AI. Decision Models is not affiliated with TypeSafe AI.