Article
What Is the OpenAI Decisions API? The Constrained-Choice Primitive Explained
Announced at DevDay 2026, the Decisions API takes context and a list of choices, then returns exactly one of them — no free text, no JSON schema, no parsing. Here is what the primitive actually is and where it fits.
The one-line definition
The Decisions API is a constrained-classification primitive from OpenAI: you send context (text, images, or both) plus a set of candidate choices, and the API returns exactly one of those choices. It is not a chat endpoint. It does not generate prose. It produces a single, guaranteed-valid answer.
What makes it different from chat
A chat model is optimized to continue a conversation. That means it can invent options that were never declared, refuse to pick, or wrap its answer in explanation. Every one of those behaviors is a bug when the downstream system needs a machine-readable decision.
The Decisions API removes the problem at the source:
- The output space is bounded by the choices you declare
- The response is a single choice, not text to parse
- No schema, no regex, no retry-and-repair loop
Where it fits in a pipeline
The API shines anywhere a system must branch on an input: customer-triage routing, moderation queues, agent tool selection, and visual verification. In each case the cost model changes too — you pay for a classification, not for a long generation.