Back to the stories

OpenAI makes Decisions API generally available with GPT-6 Luna

Score 10.0,

Developer platforms and APIs

Software can now ask an AI for a machine-ready decision instead of parsing prose and guessing what the model meant.

OpenAI made a dedicated Decisions API generally available on October 22, 2026. The endpoint uses GPT-6 Luna and accepts text, images and structured inputs, returning clear predicate answers, chosen options, or numeric scores in native JSON.

Why that matters: until now, apps had to extract decisions from ordinary model replies, adding fragile parsing, testing, and latency. Getting a decision back in a predictable format turns the model into a direct decision source you can wire into business logic and automation.

How it works: think of the API as teaching the model to reply in a form a program understands. Instead of freeform text, the model emits a structured yes or no, a single selected choice, or a confidence score the app can act on immediately. OpenAI says this route can be up to ten times faster than calling the same model through the generic Responses API.

What changes now: developers can build agents that produce actionable answers without extra glue code, and live integrations can receive decision events during conversations. The platform also adds prompt caching and options to keep inference within the United States or the European Union, which helps with latency and regulatory concerns. Importantly, OpenAI frames pricing around input and cached inputs and does not charge for the model's outputs.

What to watch next: will real-world benchmarks match the promised speed and reliability, and will businesses trust these decision endpoints for critical workflows? That will decide whether this becomes a developer standard or another optional tool.