The Decisions API is an OpenAI programming interface built on the GPT-6 Luna model that focuses the model's capabilities on questions with bounded answers. Instead of generating a full text response, it returns typed answers together with calibrated probabilities, letting applications and agents make near real-time decisions.
The API accepts text, images and native JSON and supports three kinds of outputs: predicates (estimating the probability that a statement is true), choices (selecting from a fixed set of options) and scores (scoring an input against a rubric). According to OpenAI, decisions are made up to ten times faster than using GPT-6 Luna through the Responses API.
The Decisions API became generally available (out of beta) on 22 October 2026. Capabilities include native GPT-Live integration for voice apps, prompt caching to lower costs, and support for up to 400 questions per request.
