Comparison
OpenAI Decisions API vs Structured Outputs
Structured Outputs make a model emit schema-valid JSON. The decision pattern behind the OpenAI Decisions API — and the decisions-1 endpoint on this site — does something different: it picks among a finite set of answers and tells you how sure it is about each one. Same neighborhood, different tool.
Updated
The actual difference
Structured Outputs constrain the shape of the response — the model must produce JSON matching your schema. If you ask it for a label inside that schema, you get a label, and nothing about the runner-ups.
A decision call constrains the semantics — the model picks one of your finite answers and returns a probability for every option plus a confidence value. The second-place option and its probability are the whole point: they tell you whether to act automatically or escalate to a human.
Side by side
Structured Outputs give you a parseable enum field. The decision endpoint gives you the same enum plus the distribution behind it.
Structured Outputs
// Structured Outputs: the model fills a JSON schema.
// You get a parseable label — not a probability per option.
{
"response_format": {
"type": "json_schema",
"json_schema": {
"name": "route",
"schema": {
"type": "object",
"properties": {
"team": { "enum": ["payments", "frontend", "account"] }
}
}
}
}
}The decision version
Same routing task, expressed as a choice question. The response carries the winner, the per-option probabilities, and a confidence — enough to auto-route at 0.8 and escalate below it.
Decision request
// Decision endpoint: the answer carries a probability
// for every option plus a confidence — enough to set a threshold.
{
"model": "decisions-1",
"state": "The page renders blank in Safari.",
"questions": {
"team": {
"type": "choice",
"instructions": "Which team should own this ticket?",
"criteria": {
"payments": "Checkout or billing.",
"frontend": "Rendering or browser behavior.",
"account": "Login or permissions."
}
}
}
}When to use which
Use Structured Outputs when you need the model to produce data — extraction, form filling, multi-field responses where every field matters. Use a decision endpoint when the model's job is to pick one branch and you need to know how close the call was.
They compose cleanly: a decision call picks the route, then a structured generation produces the payload for that route.
FAQ
Can Structured Outputs give me confidence scores?
Not reliably — you can ask the model to emit a confidence field, but it is a generated number, not a computed probability per option. The decision endpoint returns real per-option probabilities.
Which is cheaper?
On this endpoint a decision call is 1 credit regardless of question count (1–6 questions). A Structured Outputs call's cost depends on your provider's token pricing — compare on your own workload.
Does OpenAI's Decisions API replace Structured Outputs?
They answer different needs. Structured Outputs control response shape; a decisions API returns a constrained answer with probabilities. Per the 2026-09-29 announcement, OpenAI's version also returns probabilities — the difference stands.
See the probabilities yourself
One free trial decision for new visitors — run a choice question and read the distribution.