OpenAI released its Decisions API in public beta on October 6, giving developers a new endpoint that returns typed answers instead of free text. The API runs exclusively on the GPT-6 Luna model.

The endpoint, POST /v1/decisions, accepts text and image inputs and supports three output types: predicates that estimate a probability, choices that pick from predefined options, and scores that rate an input on a numeric range.

OpenAI says the API produces answers about 10 times faster than the Responses API, with response times around 150 milliseconds versus 1.6 seconds for a standard Luna call. It costs 0.10 dollars per million input tokens, with no output-token charges.

Use cases include content classification, customer inquiry routing, damage detection in photos and an agent's next action. The API was first introduced at DevDay 2026 in limited preview, and general availability is expected in the coming weeks.

The release puts OpenAI in direct competition with decision-model startups such as Jev, which pioneered typed decision endpoints for software.