Decision models answer predicate, choice and score questions about one input in a single request and return probabilities, which suits routing, guardrailsThe checks around a model that block bad inputs and outputs — filters, validators, and permission rules the model itself can't override.Full definition → and rubric scoring without parsing generated text.
Key takeaways · AI-distilled
On AI Gateway, openai/gpt-6-luna-decisions is a separate model ID from openai/gpt-6-luna: requests to the decisions ID go to the Decisions API and are billed at its rates, while the plain ID stays a text-generation model.
decisions.create needs OpenAI SDK 7.30.0 or later for JavaScript or 3.26.0 or later for Python. In the AI SDK, experimental_decide needs ai 7.0.128 or later, questions are keyed by name, and the yes/no type is called boolean.
Answers come back as an array in question order, each carrying its question's name, and the same request shape works with other decision models on the gateway, including Jev, Liquid d1 and Laya.
Terms in this piece · Glossary
guardrails — The checks around a model that block bad inputs and outputs — filters, validators, and permission rules the model itself can't override.
