GUIDE · UPDATED SEPTEMBER 30, 2026
What is a Decisions API?
A Decisions API is a model endpoint that does not write free text. You send context and a question with a closed list of answers, and it returns the answer it picks together with a probability for every option. The output space is fixed before the model runs, so your code only ever handles answers it already knows.
How it differs from a chat completion
A chat completion writes prose that you then parse. Even with a JSON schema, you get one sampled answer and no measure of how close the alternatives were. A Decisions API reads the model's probability for each allowed answer instead, so you can set thresholds, route low-confidence cases to a person and track drift.
| Approach | Output | Confidence |
|---|---|---|
| Chat completion | Free text to parse | None |
| JSON schema / enum | One sampled label | None, or self-reported |
| Decisions API | One label from your list | A probability for every option |
Typical jobs
Routing tickets and emails, moderating content against a policy, picking an agent's next tool, scoring sentiment or lead fit, and running yes/no guardrail checks on generated text. In each case the right answer is one item from a known list.
How this Decisions API works
Each question is shown to an OpenAI model as a multiple-choice prompt with lettered options. The model answers with one token, and we read the probability it assigns to every letter, then normalise across your options. You get choice, yes_no and score question types and can ask several questions about the same context in one request.