OpenAI unveils Decisions API, which returns predefined answers with confidence scores, available in limited preview with no pricing details

If you use AI to generate content, you can now get more predictable outputs with confidence scores.

OpenAI has launched a new Decisions API, currently in limited preview. This API is designed to provide predefined answers to queries, accompanied by confidence scores. The announcement positions this as a response to the needs of companies like TypeSafe, suggesting a move towards more structured and verifiable AI outputs. Pricing details for the Decisions API have not yet been disclosed. The API operates on a foundation referred to as 'Luna', though specifics about this underlying technology are also scarce. This development signals OpenAI's intent to offer more controlled and predictable AI interactions, moving beyond purely generative responses.

The introduction of the Decisions API signifies a strategic shift for OpenAI, moving beyond open-ended generative models to offer more deterministic AI solutions. By providing predefined answers with confidence scores, OpenAI caters to use cases where reliability and predictability are paramount, such as in enterprise decision-making processes or structured data retrieval. This approach addresses a critical need for verifiable AI outputs, potentially reducing the "hallucination" problem often associated with large language models and making AI more trustworthy for sensitive applications.

This API's availability in limited preview suggests a cautious rollout, allowing OpenAI to refine its performance and gather feedback before a broader release. The focus on predefined answers with confidence scores may impact how developers integrate AI into their applications, shifting from purely creative generation to more constrained, albeit potentially more accurate, decision support. Companies looking for AI that provides clear, score-backed recommendations will find this API particularly relevant, while those reliant on highly novel or unconstrained AI generation may see this as a more specialized tool.