OpenAI's Decisions API lets GPT‑6 Luna pick from answers you define
OpenAI's new Decisions API returns one predefined answer with a confidence score, powered by GPT-6 Luna, and is in limited preview with no published pricing.
OpenAI launched a Decisions API at DevDay on Tuesday, powered by GPT-6 Luna. In a post, OpenAI's developer account says developers define questions and possible answers, then use the model to classify content, route requests or choose an agent's next action. It is available in limited preview.
https://x.com/OpenAIDevs/status/2105003318917697873
Each request passes context, which can be text or an image, plus a closed list of options. The model returns the option it judges most likely, along with a confidence score, according to reports on the announcement. Developers do not need to parse free text or validate that the output matches an allowed value.
Reports put latency at about 150 milliseconds. That figure traces to secondary write-ups, and the paper has not found it in an OpenAI document. AlphaSignal says OpenAI describes access as limited to selected API customers, with a broader release to follow in the coming days.
OpenAI has not published pricing, rate limits, regional availability or the maximum number of choices per request, per AlphaSignal's review of the announcement. Luna is one variant in the GPT-6 family alongside Astra and Sol. The New Stack frames the API as OpenAI's answer to TypeSafe's Jev. No independent tests have been published.