A ticket arrives. Should it go to billing or technical support? For this kind of choice, generating a long response adds time before the action.
On September 29, 2026, OpenAI announced the Decisions API. It uses Luna to choose from predefined answers. It targets the same selection needs as Jev, which we covered in our Jev and Laya article.
A decision your application can use
You supply context as text or an image, a question and the possible answers. The API returns a choice your application can use to classify content, route a request or choose an agent's next action.
For example, an agent could choose between searching an FAQ, checking documents or handing over to a person. It avoids generating an explanation solely to select that step.
150 ms in OpenAI's demo
The official demonstration shows roughly 150 ms per request, compared with 1.6 seconds through the Responses API. That is about ten times faster in this demonstration, rather than a comparison with Jev or a guaranteed response time for your product.
For an agent making several consecutive decisions, shortening each step may reduce the wait. You still need to measure the complete workflow and the quality of its choices.
A Jev competitor to test on real requests
Jev already offers choices, scores and truth estimates. OpenAI confirms selection from predefined answers, with visual inputs. The two offerings are therefore not yet comparable across every use case.
As of September 30, the Decisions API is in limited preview. OpenAI plans broader access in the coming days. The official sources we reviewed do not yet specify pricing, the maximum number of choices or fine-tuning options.
In Askolia, we would start with FAQ selection: compare the cost per correct answer, response time and errors. That test will tell us which option earns its place in the application.
