OpenAI opened a public beta for a Decisions API, a new endpoint built to evaluate text and images and return fast, typed answers, according to the company's developer documentation. OpenAI said the API returns results about 10 times faster than its existing Responses API and is meant for tasks like classifying content, routing requests and prioritizing work.
The API currently runs on one model, called gpt-6-luna, through a dedicated POST /v1/decisions endpoint, the documentation says. OpenAI said it expects to reach general availability in the coming weeks.
Developers send a model specification, input evidence such as text or images, and one or more questions, then get back an array of answers, according to the docs. Questions come in three types: predicate, which checks a condition and returns a probability from 0 to 1 that it is true; choice, which picks from a fixed list of options; and score, which rates input against ordered levels.
A request can bundle multiple independent questions together, or developers can send questions in sequence when a later decision depends on an earlier one, the documentation says.
Classification and routing are workaday tasks that do not need a full chat model's price tag, and OpenAI is betting dedicated infrastructure for them, priced and sold separately, is worth shipping as its own product line. The 10x speed claim is OpenAI's own number, so developers evaluating it should benchmark it against whatever classifier they are replacing today.