How one request is structured
A request to POST /v1/decisions has three main parts:
| Field | Meaning |
|---|---|
model | Currently gpt-6-luna. |
input | The evidence to evaluate: a text string or supported user-message content. |
questions | One or more independent typed questions with instructions and options or levels. |
The three answer types
Predicate
Predicate estimates the probability that a condition is true. It is useful when the business decision can be written as one observable yes/no statement.
Choice
Choice selects one supplied value and returns a probability distribution plus confidence. Categories should be distinct, and a fallback such as “other” helps when the set is incomplete.
Score
Score evaluates ordered rubric levels. The result is a probability-weighted average of zero-based level indices, so it can fall between two named levels.
Decisions API or Structured Outputs?
| Need | Better fit |
|---|---|
| Probability of a condition | Predicate |
| One category from fixed options | Choice |
| Ordered severity or quality level | Score |
| Extract several custom fields | Structured Outputs with the Responses API |
| Generate prose or explanations | Responses API |
Current pricing and availability
As of the verification date, the API is in public beta and supports only gpt-6-luna. OpenAI lists input pricing at $0.10 per million tokens for the Decisions endpoint, with no output-token charge. Regional and long-context adjustments may apply; always check the official pricing page before relying on a cost estimate.
Practical CSV resources
- Run the Decisions API on a CSV without code
- How to batch Decisions API requests safely
- Customer feedback classification example
Sources and verification
Product facts on this page were checked against the following primary sources on October 11, 2026.