The API is built with GPT-6 Luna and is intended for bounded application and agent decisions such as classification, routing, and selecting a next action.
Developers define a question and its permitted answers, then send text, images, or native JSON. The API returns a predicate probability, a choice from supplied values with probabilities and confidence, or a score based on ordered rubric levels. OpenAI says requests can contain up to 400 questions, twice the prior limit.
The release adds native GPT-Live integration, prompt caching for reused context, and inference residency in the US and EU. OpenAI says Decisions can be up to 10x faster than using GPT-6 Luna through the Responses API. Customer examples report performance and cost results, but the announcement does not provide independent benchmarks, full pricing, or a general guarantee that those results will transfer to other workloads.