What changed
OpenAI’s October 22 changelog marks the Decisions API generally available with gpt-6-luna, currently its only supported model. The API evaluates text, images, or structured JSON and returns a typed answer: a probability, a choice from supplied options, or a score against a rubric.
Requests now support up to 400 questions, compared with 200 in public beta. The announcement also highlights shared-input prompt caching, images in individual question instructions, and direct JSON input. These are additions since the beta, rather than a claim that every capability first appeared today.
Check what the result represents
The guide says independent questions can share a request; decisions that depend on an earlier answer need separate requests. It recommends setting review thresholds using labeled examples and the costs of false positives and false negatives.
Our take
For a small business tool, a useful first trial might separate routine requests from cases needing a person. Define the categories, include an “other” route, and check real examples before letting the result drive work.
A neat label can hide uncertainty. Decide what happens when a request is ambiguous, refused, or costly to misroute, and keep the underlying evidence available for review. We have not called this API or measured its speed, accuracy, or operating cost.