OpenAI has made the Decisions API generally available, positioning it as a tool for near-real-time, structured decisioning in applications and agents. The Decisions API runs on GPT‑6 Luna and is designed to convert incoming information — text, images or native JSON — into bounded, actionable answers quickly and at lower cost. OpenAI says the endpoint can operate up to 10x faster than calling GPT‑6 Luna through the Responses API for comparable tasks, a claim aimed at teams that need rapid control-flow decisions in interactive experiences.
At its core, the Decisions API asks developers to define a question, enumerate the allowed answers, and provide relevant context. That context can include plain text, images embedded in instructions, or structured JSON. For each request the service returns one of three answer types — predicate, choice or score — enabling applications to determine a next step in a workflow rather than receive freeform text.
The three answer types have distinct roles. A predicate returns an estimated probability between 0 and 1 that a condition is true (for example, whether a product shows visible damage). A choice selects one predefined option and includes probabilities and a confidence measure, which is useful for routing or control-flow decisions such as continue, retry or clarification. A score produces a probability-weighted average mapped to ordered rubric levels, suitable for graded outputs like severity ratings or other ordinal assessments.
Developers can tune the Decisions API to many use cases by choosing the appropriate answer type and supplying targeted context. The service supports up to 400 questions per request — double the previous limit — and accepts images in instructions as well as native JSON inputs, allowing multimedia and structured triggers to feed decision logic in a single call.
OpenAI highlights internal and customer examples to illustrate practical uses. Internal product teams have used the Decisions API for transaction categorization in Personal Finance, theme classification and narrative scoring in Communications, and interactive product features. Customers have applied Decisions to tasks such as attributing AI costs and building more interactive language-learning experiences. These examples emphasize how fast, structured decision-making can turn incoming data into actionable outcomes within product and business workflows.
The Decisions API also integrates with GPT‑Live to enable real-time voice and conversational experiences. In live sessions, Decisions can evaluate transcript updates and emit decision events that an application can act on immediately — for example, deciding whether to ask for clarification, route a request, or escalate an issue. OpenAI demonstrates avatar reactions in the documentation: live decision events emit a decision.event and the client uses the selected reaction to animate expressions during conversation. A time-limited demo and developer documentation show how to add decision triggers to GPT‑Live session configurations and respond to decision events in real time.
Performance and regional processing options were expanded since the public beta. OpenAI added prompt caching for reused context, extended inference residency to the US and EU, and reported continued latency improvements along with added capacity in several regions in Asia. For customers with stricter data requirements, Zero Data Retention is available to eligible customers and regional processing options are listed for US and EU residency needs.
On pricing and availability, the Decisions API is generally available today on a dedicated endpoint using GPT‑6 Luna. OpenAI lists input-token pricing at $0.10 per million tokens, with cached input charged at $0.01 per million tokens. Cached writes are available with 30-minute and 12-hour durations. OpenAI notes that customers pay only for input tokens and that there are no output-token charges. The company points developers to the Decisions documentation and a decisions playground as recommended starting points for testing decision examples that reflect real usage.
For teams building interactive, low-latency experiences, the Decisions API offers a structured alternative to open-ended generation: define the decision space, supply context, and receive a bounded answer type that maps directly to application logic. As OpenAI makes the endpoint broadly available and adds regional and caching options, the Decisions API becomes a focused option for applications and agents that need predictable, fast decisioning across text, images and structured inputs.
More information, including documentation, pricing and the developer guides, is hosted by OpenAI at the Decisions API landing page.
Source: Read the original source
Leave a Reply