Decisions API
OpenAI's announced decision layer: developers pose a question with a fixed set of possible answers plus text or image context, and the model returns a choice, for classification, routing or an agent's next action.
The Decisions API is a product OpenAI announced at DevDay 2026 as a real-time decision layer. As described in the launch material, developers define a question with finite, pre-defined answers, supply text or image context, and use the result for tasks such as classification, request routing or choosing an agent's next action. At the time of writing it was in limited preview, and public documentation did not yet specify a dedicated endpoint, pricing or a full confidence contract.
Why it matters
It moves model judgment closer to action. When a model returns a choice instead of prose, something must decide when that choice is trusted enough to act on. That is the design problem: threshold policy, calibration, authority and recovery matter more than the answer itself. A decision layer also creates a new kind of dependency, because a vendor's decision behaviour can end up embedded in many operational thresholds.
Reported details such as response speed and confidence scores should be treated as unconfirmed until the production contract is public.
Read more in OpenAI Decisions API turns probability into policy.
Related terms
Decision layer
A dedicated component between AI judgment and organizational action that decides whether the work should continue, separate from the part of the system that does the work.
Threshold policy
The explicit rules that turn a model's probability or score into an action — reject, send to a human, or act autonomously — owned and reviewed separately from the model itself.
Calibration
How well a model's stated confidence matches reality: if it says 90% on a hundred comparable cases, roughly ninety should be correct. Once probabilities drive actions, it becomes an operating metric.
Confidence score
A number a model or classifier reports to indicate how sure it is. It is only useful for decisions if it is calibrated, and it never decides on its own what action is permitted.