Decision model
A decision model is a model built to return a typed answer, such as a choice from a list, a score on a rubric or a yes/no probability, instead of generating free text.
By Divya Sharma · Checked Nov 15, 2026 · How we review
Agents spend much of their time on small judgements: which team owns this ticket, is this message spam, should this step need human approval. A chat model can answer these, but it replies in prose that code has to parse, and it gives no dependable measure of how sure it is. A decision model is built for that narrow job.
How it works
You send the model the evidence (a piece of text, JSON or an image) and one or more typed questions. TypeSafe, which calls its Jev model a "System One" model, describes three question types: a choice from a list of options, a score on a rubric, and the probability that a statement is true. The answer comes back as values with probabilities, so code can compare them against a threshold and act, or pass the case to a person when confidence is low. Other examples include Laya and GLiNER2.5-Decide.
Where it fits
Typical uses are LLM routing, triage, guard checks before a tool call, and gating whether an agent continues. It sits alongside guardrails and structured output: structured output makes a text model fill in a schema, while a decision model is designed around the typed answer from the start.
Limits
The labels are read from the request rather than trained in, which makes them flexible, but accuracy on your categories is not guaranteed. Test a decision model on a labelled sample of your own data before trusting its scores.
Related terms
- LLM routing
- LLM routing is the practice of sending each request to the language model best suited to it, often a cheap model for easy prompts and a stronger one for hard prompts, to balance cost, speed and quality.
- Structured output
- Structured output is an API feature that makes a language model's reply follow a format you define, usually a JSON schema, so that code can parse it reliably instead of scraping free text.
- Guardrails (AI)
- Guardrails are runtime checks placed around a language model that inspect inputs and outputs and block, rewrite or redirect them, so an application stays on topic, avoids harmful content and follows rules the model alone would not reliably keep.
- Small language model
- A small language model (SLM) is a language model with relatively few parameters, usually a few billion or fewer, built to run on a laptop, phone or single GPU at lower cost than a large model.
- LLM-as-a-judge
- LLM-as-a-judge is an evaluation method where a strong language model grades the outputs of another model or application, using a rubric, a reference answer or a side-by-side comparison, in place of or alongside human reviewers.