News
What Is the OpenAI Decisions API? Fast AI Choices, Explained
The OpenAI Decisions API returns a yes/no, a pick or a score instead of text. Now generally available on GPT-6 Luna at $0.10 per million input tokens.
The OpenAI Decisions API is a new way to ask an AI model a question when you do not want an essay back. You hand it some text or an image, list the possible answers, and it tells you which one fits and how sure it is. OpenAI made it generally available on October 22, 2026.
If you have seen headlines saying OpenAI is moving onto Jev’s turf, this is the product they mean. Here is what it does, what it costs, how it stacks up against Jev, and why a tool for developers matters to anyone who uses AI agents.
What is the OpenAI Decisions API?
Most AI calls work like a conversation. You send a prompt, and the model writes a reply word by word. That is great for drafting an email. It is slow and wasteful when all you need is "billing" or "shipping."
The Decisions API skips the writing. A developer sends a request to one endpoint, POST /v1/decisions, with an input and a list of questions. Each question has a type, and the answer comes back as a number or a label that code can use right away. Nothing rambles, and nothing needs to be parsed out of a paragraph.
The three question types, in plain English:
| Type | What it returns | Example from OpenAI’s docs |
|---|---|---|
| Predicate | A probability from 0 to 1 that a statement is true | Does this product photo show visible damage? |
| Choice | One option from your list, with probabilities and a confidence value | Route this complaint to billing, technical, shipping or other |
| Score | A rating on 2 to 10 ordered levels, as a weighted average | How severe is this issue, from "workaround available" to "fully blocked"? |
A score can land between two levels. If the result is 1.1, the issue sits a little past the second level on the scale. That is useful, because it tells software how close a case is to the line.
When did it come out?
The rollout took about three weeks:
- September 29, 2026: Sam Altman mentioned it at OpenAI DevDay as a limited preview. TechCrunch quoted him saying that focusing the model on one choice lets OpenAI make it "extremely fast."
- October 6, 2026: OpenAI opened a public beta to all developers.
- October 22, 2026: General availability, with new features added since the beta.
The additions at launch include prompt caching, native JSON input, images inside individual questions, and an integration with GPT-Live, OpenAI’s voice model, so voice apps can react while a person is still talking. The limit also doubled from 200 to 400 questions per request.
OpenAI Decisions API cost
This is the part that makes developers pay attention. You pay only for what you send in. The answers are free.
| Item | Price |
|---|---|
| Input tokens (not cached) | $0.10 per 1M tokens |
| Cached input tokens | $0.01 per 1M tokens |
| Output tokens | Not charged |
| Supported model | gpt-6-luna only |
To make that concrete: a million tokens is very roughly 750,000 words. A short support ticket might be a hundred tokens, so a dollar covers tens of thousands of routing decisions. OpenAI notes that regional processing premiums and long-context multipliers can apply, and cache writes are billed per its pricing guide, so real bills can run a bit higher than the headline number.
OpenAI Decisions API vs Jev
Jev, from San Francisco startup TypeSafe AI, came first. It launched in early access on September 15, 2026, and does the same basic job: a yes or no, a choice, or a score, each with a confidence number. We explained it in What is Jev AI? When OpenAI showed its version two weeks later, TypeSafe’s CEO Diogo Almeida joked on X about the "clone wars."
| OpenAI Decisions API | TypeSafe Jev | |
|---|---|---|
| Answer types | Predicate, choice, score | Yes/no, choice, score |
| Input price | $0.10 per 1M tokens | $0.042 per 1M tokens (listed by TypeSafe) |
| Output price | Free | Free |
| Images | Yes | Not offered, per early comparisons |
| Status | Generally available, Oct 22 | Available to developers |
So OpenAI costs about 2.4 times as much per input token, but it takes images and sits inside the OpenAI account many companies already have. Microsoft’s Decision-1 and Inception’s Mercury Decide target the same space. Which one is most accurate is still an open question. Early comparisons come from small blog tests, not independent benchmarks, so treat any "winner" claims carefully.
Why AI agents need a decision model
Here is the bigger reason this matters. An AI agent makes hundreds of tiny choices while it works. Which tool should it call next? Is this web page safe to act on? Should a human check this before it goes out?
Asking a full writing model each of those questions is slow and costly. A decision endpoint answers in a fraction of the time, so an agent can check every single action against its task before taking it. TechCrunch described exactly this use: a cheap watcher that blocks risky agent actions, flags unclear ones for review and lets the rest through. OpenAI’s own docs list routing requests, flagging images for review and grading conversations that include tool calls.
In short, decision models are becoming the reflexes of agent systems, while big models stay the brains.
What this means for you
You will not open the Decisions API on your phone. It is plumbing for developers. But plumbing shapes what you feel: faster agents, fewer wrong turns, and lower costs that make AI tools cheaper to offer.
The practical answer for most people is to use an app that already handles this layer. Agentic AI gives you a team of expert agents on iPhone and iPad, running on frontier models, so you describe the job and the agents sort out the steps. There is no account to create, and your first 3 tasks are free. You can learn the vocabulary in our agentic AI glossary or start with what agentic AI is.
Quick answers
What is the OpenAI Decisions API? An OpenAI endpoint that answers closed questions with a probability, a choice from your list, or a score, instead of generating text. It runs on GPT-6 Luna.
How much does the OpenAI Decisions API cost? $0.10 per million input tokens, $0.01 per million cached input tokens, and nothing for output. Regional and long-context surcharges can apply.
Is the Decisions API better than Jev? It is not proven either way. OpenAI’s version accepts images and is part of the main OpenAI platform; Jev is cheaper per input token. Independent accuracy tests are still thin.
Is the OpenAI Decisions API available now? Yes. It became generally available to developers on October 22, 2026, after a public beta that began October 6.
Can I use the Decisions API in ChatGPT? No. It is a developer API. Regular users feel it indirectly, through apps and agents that use it behind the scenes.
Sources: OpenAI Decisions guide · OpenAI Developer Community: GA announcement · OpenAI Developer Community: public beta · Unite.AI · TechCrunch · Crypto Briefing · Jev pricing
