The OpenAI Decisions API is a developer tool, made generally available on 22 October 2026, that answers questions with bounded answers very quickly: is this true, which of these options, or how severe on a scale. It's built with GPT-6 Luna and OpenAI says it can make near real-time decisions from text, images and structured data up to 10 times faster than GPT-6 Luna through the Responses API. You won't use it in the ChatGPT app; it's for the apps, chatbots and automations your developer builds.
Most AI news is about bigger, smarter models. This one is the opposite. It's about the hundreds of tiny calls inside a business system: sorting an email, routing a support request, deciding whether a photo shows damage. Each one is simple, but OpenAI's point is that at scale, the waiting adds up.
If you run a business and someone has mentioned this to you, here's the plain-English version, and when it's worth a conversation with your developer.
The three kinds of answer it gives
Developers set a question and the answers allowed, then pass in the context as text, images or JSON. The API replies with one of three answer types:
| Type | Use it to | What comes back |
|---|---|---|
| Predicate | Estimate whether a condition is true, such as whether a product has visible damage | A probability between 0 and 1 |
| Choice | Pick one of your own options, such as continue, retry or ask for clarification | One of your options, with probabilities and confidence |
| Score | Rate something against ordered levels in a rubric, such as damage severity | A probability-weighted average of the levels |
The useful part is that these are structured answers your system can act on straight away, rather than a paragraph someone has to read.
What's new at general availability
OpenAI says it has added several things since the public beta:
- A native integration with GPT-Live, its voice model, so decisions can happen during live conversations
- Prompt caching for context you reuse
- Inference residency in the US and EU
- Native JSON inputs and images inside question instructions
- Up to 400 questions per request, twice the previous limit
- Lower latency, including extra capacity in several regions in Asia
Real uses, in business terms
OpenAI shared how its own teams use it. Its Personal Finance team uses it to categorise transaction descriptions. Its Communications team uses it to classify press coverage by theme and score whether a piece reinforces or challenges key messages, so people know what needs a closer look.
Translate that to a small business and you get jobs like these:
- Inbox triage. Is this enquiry a sales lead, a support question or spam?
- Support routing. Should the chatbot answer, ask a clarifying question, or hand over to a person? OpenAI gives this exact example for voice agents.
- Quality checks. Does this product photo meet our listing rules?
- Review sorting. Is this review positive, neutral or a complaint that needs a reply today?
If any of these already happen inside your tools with AI, the Decisions API is the kind of thing that can make them faster.
Do you need it?
Probably not directly. If you use ChatGPT, Claude or Gemini in the normal apps, nothing changes for you. The Decisions API matters if:
- You have, or are planning, a custom chatbot, voice agent or automation that makes lots of small yes-or-no or pick-one decisions.
- Speed matters, for example a support agent that shouldn't pause before routing a caller.
- You handle high volume, where many small calls add up.
If that's you, it's a conversation to have with whoever builds your systems. If you're not sure what the OpenAI API is in the first place, my explainer on what the OpenAI API is covers the basics.
Questions to ask your developer
Copy these questions
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Which decisions in our system are bounded: yes or no, pick one, or a score?
Would moving those to the Decisions API make the customer-facing parts faster?
Where should our data be processed, and does US or EU residency matter for us?
How will we check the answers are right before we rely on them?
That last one is the most important. A fast answer is only useful if it's the right one, so any system like this needs a human check while it beds in, and a way to flag low-confidence results. The choice answer type returning probabilities and confidence is what makes that possible.
What it doesn't do
It's worth being clear about the limits of what OpenAI announced. The Decisions API answers bounded questions; it isn't for writing replies, summarising documents or holding a conversation, which is what the main models are for. It also doesn't decide what your rules should be. You still define the question, the allowed answers and what your system does with each one.
Where this fits in the bigger picture
This is part of a wider shift towards AI agents that do the routine sorting and routing work in the background. My guide to how AI agents work for business owners explains that shift without the jargon.
My verdict: not a tool for most owners to touch, but a useful signal. The plumbing behind AI customer service is getting faster, which means the experience your customers get from a well-built agent is about to improve. If you'd like help working out whether an agent makes sense for your business, that's what I do in my one-to-one AI sessions.
Frequently asked questions
What is the OpenAI Decisions API?
It's a developer tool, generally available from 22 October 2026, that answers questions with bounded answers, such as a probability, a choice from your options or a score, from text, images and structured inputs.
How fast is the Decisions API?
OpenAI says it can make decisions up to 10 times faster than GPT-6 Luna through the Responses API.
What answer types does the Decisions API support?
Three: predicate, which returns a probability that a condition is true; choice, which picks one of your options; and score, which rates an input against ordered levels.
How many questions can I send per request?
OpenAI says up to 400 questions per request, twice the previous limit.
Can I use the Decisions API in ChatGPT?
No. It's an API for developers building apps, agents and automations, not a feature inside the ChatGPT app.
