OpenAI Officially Launches Decisions API in the United States

en.Wedoany.com Reported - On October 22, the U.S. artificial intelligence company OpenAI announced that its Decisions Application Programming Interface (Decisions API) has concluded public testing and is now officially and fully available, powered by the GPT-6 Luna model. The interface is designed for real-time decision-making tasks for applications and AI agents. Input is priced at $0.10 per million tokens, cached input at $0.01 per million tokens, and no output token fees are charged.

The Decisions API can process text, images, and structured data. Developers can predefine questions and candidate answers, and the interface returns results in three categories: probabilities, options, or scores, for use in content classification, request routing, task prioritization, and automated workflow control. Performance data released by OpenAI shows that, compared with performing the same types of tasks through the GPT-6 Luna Responses API, the Decisions API can deliver response speeds up to 10 times faster. A single request supports up to 400 questions, double the previous limit.

The generally available version adds native integration with the GPT-Live real-time voice system, enabling applications to invoke decision-making functions during user voice interactions. The interface also adds prompt caching, native JSON input, and image instruction support, and offers regional inference processing services in the United States and Europe. Eligible customers can use the zero data retention feature.

OpenAI has already applied the interface to internal products and business processes. Among them, the Dots team used the Decisions API to shorten response times, the personal finance team used it for transaction description classification, and the communications team used the interface to classify and score media coverage by topic. Among external companies, Box used it for business document classification and workflow assignment, achieving roughly a 5x increase in task routing speed across 30 insurance certificate test cases; Ramp reported a 4x increase in the processing speed of AI cost attribution and a 30% reduction in costs.