OpenAI said its Decisions API is in public beta. A request asks one or more questions about text, images, or both. Each question gets one typed answer: the chance a condition is true, a pick from a fixed list, or a score on ordered levels. OpenAI says it runs about 10 times faster than the Responses API. Only gpt-6-luna works with it, at $0.10 per 1M input tokens and no output charge.
Choice and score answers carry a probability for each option, plus a confidence field. A score averages the level numbers, weighted by probability, so 0.1, 0.7 and 0.2 across levels 0, 1 and 2 give 1.1. OpenAI expects general availability in the coming weeks but gave no date. OpenAI's 10x speed claim comes with no independent test cited. Regional processing and long inputs cost extra, and the amounts were not stated.
Because the API returns odds instead of a verdict, the buyer sets the cut-off. OpenAI advises using your own labelled examples to pick thresholds, weighing false alarms against missed cases. Before committing, run a labelled sample, and add a fallback choice such as "other" so unclear cases go to a general review queue. Billing covers input only, so cost follows what you send, not what the model writes.