OpenAI Opens Decisions API As Public Beta With GPT-6 Luna Only Model
This is an infrastructure play: OpenAI is carving out a narrow, high-speed inference layer from the general-purpose Responses API, and the only model exposed at launch is GPT-6 Luna, which ties the product's cost profile to one model tier.
Reporting from 1 source: GIGAZINE.
OpenAI released a public beta of the Decisions API, a judgment-only endpoint that accepts text and image input and returns probabilities, selected results, and confidence scores. Developers call it through a dedicated POST /v1/decisions endpoint. The article states the only model available at launch is GPT-6 Luna, with general availability planned within a few weeks.
The Decisions API accepts text and images and returns machine-parseable judgments instead of prose. The article names three formats: predicate, which returns a probability that a condition holds; choice, which picks among developer-supplied candidates; and score, which grades input on a predefined scale.
The article gives the example of a double-charged order routed among billing, technical support, shipping, or other. Confidence values come back with the answer, so developers can send low-confidence cases to human review. Scores can land between grades, like 1.1 across 0, 1, and 2.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.