OpenAI’s Decisions API has been in public beta for some time. Now, the tech giant led by Sam Altman has announced that the Decisions API is officially available to users. This specialized model interface enables developers to build AI agents and applications, allowing software to answer recurring questions and make simple decisions. The Decisions API accepts inputs—such as images, text, and structured JSON data—provided by developers. It is powered by OpenAI’s GPT-6 Luna AI model. The company also stated that the Decisions API now features native integration with GPT-Live. Additionally, since the launch of the public beta, the company has introduced prompt caching in select markets, allowing for the reuse of contextual information.
Decisions API: Pricing and Features
In a blog post on Thursday, OpenAI announced that the Decisions API is now officially available to developers in select markets worldwide. The API can be accessed via the Decisions Playground. This specialized interface, based on the GPT-6 Luna model, is priced at $0.10 (approximately 10 rupees) per million input tokens. Meanwhile, cached input is priced at $0.01 (approximately 1 rupee) per million tokens. The Decisions API is also available to eligible customers under a “zero data retention” policy.
Developers can leverage the GPT-6 Luna-based Decisions API to assist their AI agents and applications in making small, repetitive decisions—such as classifying and organizing information, routing requests, or determining the next course of action. The AI giant claims that the Decisions API can facilitate “near real-time” decision-making using images, text, and structured inputs, operating 10 times faster than the GPT-6 Luna model when used via the standard Responses API.
Furthermore, OpenAI states that developers can provide native JSON input and include images within their prompt instructions, while also being able to pose up to 400 questions per request via the Decisions API. Since the public beta launch of this specialized model, this limit has been increased from 200 to 400 questions per request. In addition to raising this limit, the company has introduced native GPT-Live integration, prompt caching for context reuse, and the option to host inference in the United States or the European Union.
Developers provide the Decisions API with a question, a set of relevant answers, and contextual information; the API then selects the most appropriate answer based on that context. Consequently, it can deliver actionable results for applications or AI agents. Through its native integration with GPT-Live, the Decisions API can now assist applications, voice agents, and other AI-powered customer service representatives in determining whether to ask clarifying questions, redirect requests, or escalate specific issues.