OpenAI Previews Decisions API for Faster, Constrained Model Choices

At OpenAI’s Dev Day on Tuesday, CEO Sam Altman revealed a new “Decisions API,” describing it as a way to give the Luna model a predefined set of options. [1] Examples included image-classification categories and different agent behaviors. [1]

Altman said narrowing the model to a choice can make it extremely fast while retaining image understanding, broad language support, and safety protections. [1] OpenAI released the API in limited preview, and TechCrunch had not yet seen developers test it publicly. [1]

TechCrunch said the API apparently offers functionality similar to Jev, a TypeSafe AI model released earlier that month and explicitly designed for software automation. [1] Jev lets developers supply a set of choices and returns those choices as probabilities cheaply and at high speeds. [1]

TypeSafe CEO Diogo Almeida called OpenAI’s interest a possible sign that building in a “System One compatible way” is the future. [1] TechCrunch noted that the degree of similarity remains unclear because the Decisions API is still in limited preview. [1]

The Decisions API puts a constrained-choice interface around Luna, favoring predefined options over unconstrained generation when software needs classification or agent-behavior selection. Jev is the direct competitive reference in the available reporting: both focus on selecting among supplied choices, but Jev is already described as being used by developers.

Because OpenAI’s release remains limited and has not been publicly tested, claims about practical equivalence should remain provisional. What the sources don't address: how the Decisions API will be priced, how its latency and accuracy compare with Jev, or when access will broaden.