OpenAI answers TypeSafe's Jev with a Decisions API built on Luna

OpenAI used its annual DevDay conference on Tuesday to announce a Decisions API, its answer to the decision models that TypeSafe's Jev has made popular in recent months. The new API is built on Luna, the smallest and least expensive model in OpenAI's current lineup.

What the Decisions API does

Unlike chat models, a decision model is not designed to write prose. It returns one of a set of predefined answers together with a confidence score. The developer supplies the question, the candidate answers and the context, and gets back a score that can be used directly in application logic.

OpenAI says its model returns results in 150 milliseconds, where the same task would take GPT-6 Luna 1.6 seconds. That speed, together with more realistic confidence scores, is what makes decision models useful for classifying content, routing requests or picking an agent's next action from a limited set of choices.

Between a prompt and a trained classifier

Most teams handle this today with a chat model and a carefully worded prompt, asking it to pick from a list and then reading token probabilities to approximate a confidence score. The alternative is training a small classifier, which is fast and cheap but needs labelled data and a retraining run every time the label set changes. A decision model sits between the two: new labels go into the prompt, and the answer comes back as a score a developer can work with.

The idea is not entirely new for OpenAI. Its Moderation API has long returned per-category scores rather than prose, though there the categories are set by OpenAI and not by the developer.

Limited preview, open questions

For now the Decisions API is available in limited preview, with broad availability promised for the coming days. An OpenAI spokesperson told The New Stack that the company plans to share more details "at broad rollout".

What remains unclear is what a call will cost, how many candidate answers a single request can carry, and whether developers will be able to tune the model on their own data. Those answers will decide whether the Decisions API becomes a standard building block in agent frameworks or stays a niche tool beside the chat models.