OpenAI Launches Decisions API for Real-Time Classification and Routing
At its 2026 Developer Day on September 30, OpenAI announced the launch of Decisions API, a new interface designed for real-time, low-latency classification and routing. The API returns results in roughly 150 milliseconds, about 10x faster than using Luna through the regular API.
Positioned for rapid single-shot decision-making, Decisions API currently builds on OpenAI's small model Luna, focusing the model's capabilities on a set of user-defined questions with a limited set of allowed answers. Developers provide context (text or images) and predefined question-answer options, and the API returns a structured decision, including the selected answer and confidence scores, for use in downstream business logic.
Among similar products is TypeSafe's Jev, but Decisions API specifically targets turning model intelligence into fast decisions.
Typical use cases include automatically classifying issue type, priority, and assignment in customer service ticketing; selecting among predefined policies such as allow, rate-limit, or human review in content moderation; serving as a decision node in agent orchestration to choose the next tool or sub-agent; and assisting real-time business rule engines by recommending decisions in ambiguous cases.
