OpenAI makes Decisions API generally available, built on GPT-6 Luna

What the Decisions API does

OpenAI describes the problem simply: applications and agents make many small decisions in a workflow, such as how to classify an input, where to route a request, or which action to take next. Each one is simple, but at scale latency and cost add up.

The Decisions API focuses the intelligence of GPT-6 Luna on questions with bounded answers. Developers define a question and its allowed answers, then provide context as text, images, or native JSON. The API returns one of three answer types that an application can use to decide what happens next.

• predicate: estimates the probability that a condition is true, for example whether a product has visible damage. The result is a value from 0 to 1.
• choice: selects one of the developer's predefined options, such as whether an agent should continue, retry, or ask for clarification. The response also includes probabilities and confidence.
• score: rates an input against ordered levels in a rubric, such as the severity of damage. The score is the probability-weighted average of the level indices.

New since the public beta

OpenAI lists several additions since the public beta. These include a native GPT-Live integration, prompt caching for reused context, and inference residency in the US and EU. Developers can pass native JSON inputs, include images in question instructions, and ask up to 400 questions per request, twice the previous limit. The company says it has also continued to improve latency, including additional capacity in several regions in Asia.

The GPT-Live integration is aimed at voice applications. According to OpenAI, a customer support agent can use it to decide whether to ask a clarifying question, route a request, or escalate an issue. An avatar can also change its facial expressions as a user speaks, responding with a wink or a smile based on the user's words. In the example OpenAI gives, developers add a configuration to session.triggers in an existing GPT-Live session. Live emits a decision.event, and the application reads the selected reaction and animates the character accordingly.

OpenAI also offers a time-limited demo in which the API chooses a facial expression based on what the user and the assistant say.

How teams are using it

OpenAI says teams across the company already apply the API to product and business workflows. The dots team uses it to reduce response times and improve interactivity. The Personal Finance team uses it to categorize transaction descriptions. The Communications team uses it to classify coverage by theme and to score whether it reinforces or challenges key narratives, which helps identify what warrants closer review. The company adds that customers use the API for tasks ranging from accurately attributing AI costs to making language-learning experiences more interactive.

Pricing, availability and getting started

The Decisions API is available through a dedicated Decisions endpoint with GPT-6 Luna. Pricing is $0.10 per million input tokens, and cached input costs $0.01 per million tokens. OpenAI says developers pay only for input tokens, with no output-token charges. Details on 30-minute and 12-hour cached writes are in the company's pricing guide.

Regional processing is available in the US and EU, and Zero Data Retention is available to eligible customers.

OpenAI advises starting with one recurring decision that an application makes often: define the question, the possible answers, and the context needed to tell them apart. Developers can then test it on realistic examples in the Decisions Playground with their own text or images, or follow the developer guide to make a first request.

What it means for developers

For teams that run many small classification, routing or next-step decisions, the release offers a dedicated endpoint with input-only pricing and structured outputs. Regional processing in the US and EU and Zero Data Retention for eligible customers address data-handling requirements, while the GPT-Live integration extends the same approach to live voice conversations.