OpenAI's Decisions API lands on AI Gateway: swap the base URL and model ID to ask for probabilities, choices and scores
Synopsis
OpenAI's Decisions API is now callable through AI Gateway at /v1/decisions, using the same request and response format as OpenAI, so existing OpenAI SDK integrations can switch by changing the base URL, credentials and model ID. Decision models answer typed questions about shared input and return structured answers rather than generated text, supporting predicate, choice and score questions for tasks such as classification, routing, triage and rubric-based scoring. The endpoint works with any decision model on AI Gateway, including Jev, Liquid d1 and Laya; OpenAI's GPT-6 Luna Decisions is reached as openai/gpt-6-luna-decisions, separate from the text-generating openai/gpt-6-luna.
Interpretation
OpenAI's Decisions API can now be called from AI Gateway at the /v1/decisions path, so developers no longer have to integrate directly against OpenAI's endpoint. The endpoint keeps the same request and response format as OpenAI, so an existing OpenAI SDK integration switches over by changing only the base URL, credentials and model ID. The release states this compatibility and includes JavaScript and Python examples that point the client at AI Gateway and call decisions.create.
Decision models answer typed questions about shared input and return structured answers instead of generated text. One request can carry three kinds of question: a predicate question estimates the probability that a condition holds, a choice question selects from supplied options, and a score question rates the input against an ordered rubric. The answers target tasks such as classification, routing, triage and rubric-based scoring, and the response returns answers in question order, each carrying its question's name.
The endpoint works with any decision model on AI Gateway, including Jev, Liquid d1 and Laya. OpenAI's own GPT-6 Luna Decisions is called as openai/gpt-6-luna-decisions, a different model ID from the text-generating openai/gpt-6-luna. Calling it requires OpenAI SDK 7.30.0 or later for JavaScript or 3.26.0 or later for Python.
Perspective
The demonstrated scope is the availability of the /v1/decisions endpoint on AI Gateway, its OpenAI-compatible request and response format, the three question types, and the callability of models including Jev, Liquid d1, Laya and openai/gpt-6-luna-decisions. The intended readers are developers already using the OpenAI SDK or AI Gateway, and the examples cover JavaScript and Python clients.
The release reports no accuracy or comparison results for these decision models on classification, routing or scoring tasks, and does not describe capability differences among them; that answers return in question order with their question names is the structural convention it does state.