OpenAI has introduced a Decisions API for applications that need a classification rather than a written response. It accepts text, images or both and can return yes-or-no probabilities, a choice among predefined categories, or a rating on a scale.
OpenAI says the service runs about 10 times faster than its Responses API. Suggested uses include sorting customer questions, identifying damage in photos and classifying documents. The public beta currently supports only the gpt-6-luna model, priced at $0.10 per million input tokens with no charge for output tokens. General availability is planned but not dated.
The API supports zero-data-retention configurations and HIPAA-eligible use in the United States and Europe, according to the company. Narrow outputs can simplify systems that otherwise prompt a general model and parse its prose, while probabilities make thresholds explicit. The limitation is equally clear: a fast score is still model output and needs validation for the specific images or text it will judge. OpenAI also consolidated paid API accounts into Build, Launch and Grow tiers, with monthly usage limits of $500, $5,000 and $200,000.