Updates·October 7, 2026, 00:35

OpenAI Decisions API: Fast answers with GPT-6 Luna

AI-generated and checked against the sources listed below.

OpenAI has opened the Decisions API in public beta: a fast service for developers that sorts, routes and evaluates text and images with GPT-6 Luna.

AI-generated image

What's new

On October 6, 2026, OpenAI opened the Decisions API in public beta. It is a new service for developers who build AI into apps and systems. Previously, it was only in limited preview. OpenAI expects to make it generally available in a few weeks.

The idea is simple. Instead of asking an AI to write a long text, a developer asks a question and provides a fixed list of possible answers. The AI then picks one of them. It all runs via a new endpoint (a technical address), /v1/decisions, and uses the model GPT-6 Luna. It can read both text and images.

What's clever

The service can ask three kinds of questions, and there can be several in the same request:

- Yes/no questions (predicate): the answer is a probability between 0 and 1 that something is true. - Choice: the AI picks one option from a fixed list and states how confident it is. - Score: the AI rates the input on a ranked scale, for example how serious something is.

Because the answers are fixed and structured, a company's system can use them directly without interpreting text. OpenAI says the Decisions API is about 10 times faster than the regular Responses API. OpenAI does not give a specific figure in milliseconds in the documentation. One independent outlet mentions around 150 milliseconds versus 1.6 seconds, but that has not been independently measured.

There are also some limitations. Images must be sent directly as base64 data, not as a link or file ID. And if one choice depends on another, they must be sent in separate requests. Right now, only the model GPT-6 Luna is available.

Cheaper or better?

The price is $0.10 per million tokens (pieces of text) of input. There is no charge for output and none for reading from or writing to the cache (an intermediate storage feature). Surcharges do apply for processing in specific regions and for very long inputs.

By comparison, according to third-party sources, GPT-6 Luna in the regular API costs $0.10 per million tokens of input and $0.50 of output. The Decisions API thus ends up cheaper, because output is free and the answers are short to begin with. There is no comparison with competitors' equivalent services in the sources, so we cannot provide one. The service is for developers and businesses. There is no monthly price for ordinary users and nothing disclosed about special terms for Denmark or the EU.

What it's good for

The Decisions API suits tasks where many small, quick decisions need to be made:

- Sorting content, for example whether a comment is inappropriate or whether an email is spam. - Routing inquiries, so a customer support case lands with the right department. - Prioritizing work, for example which cases are most urgent. - Choosing an AI agent's next step.

For ordinary users, it is felt mostly indirectly, through apps responding faster. It is not made for writing texts or holding conversations. It is still a beta, so details may change when it is launched broadly.

Sources

More on this topic

Get the week's AI news in your inbox

Choose your level, topics and length. One email a week, unsubscribe at any time.

Subscribe to Promptly Newsletter
PromptlyNewsletterRSSLog in

The news on aijour is AI-generated and checked against the cited sources.