OpenAI has released the Decisions API in public beta, offering a dedicated endpoint for rapid evaluation of text and image inputs. According to the official documentation, the service is currently limited to the gpt-6-luna model, with a general availability launch expected in the coming weeks.
This image shows the OpenAI Decisions API logo, which corresponds to the API discussed in the first paragraph detailing its public beta launch and gpt-6-luna model limitation.
Each request specifies the model, input (text or user messages containing text and images), and one or more questions to evaluate. Developers can choose predicate questions that return a probability estimate (0–1) for a condition, choice questions that select a category from a predefined list, or score questions that rate an input against ordered levels and output a probability‑weighted average. The API returns an answers array where each answer includes a unique name, type, and the relevant result, enabling multimodal and multi‑question workflows.
A community‑maintained LLM plugin simplifies usage from the command line, allowing natural‑language queries such as 'Does this message explicitly request a refund?' or 'How severe is this issue?' and returning structured JSON with probabilities, choices, or scores. This plugin demonstrates how developers can integrate the Decisions API with existing tooling while the service remains in beta.
OpenAI states that the Decisions API delivers typed answers approximately ten times faster than the existing Responses API, addressing latency‑sensitive use cases such as real‑time content moderation or routing. This performance improvement is highlighted in the documentation as a key motivation for the dedicated endpoint.