OpenAI Opens Its Decisions API: Pricing and Uses

On October 22, 2026, OpenAI made its Decisions API available to all developers. It is a service that doesn't write answers: it takes a question with fixed answers and returns the chosen option, a probability or a score. It matters to anyone who wants to sort messages, filter content or steer a chatbot without paying for a large model.

What we know

  • What it is: according to OpenAI's announcement, the Decisions API is built with GPT-6 Luna and focuses on questions with bounded answers, over text, images or JSON data. OpenAI says it responds up to 10 times faster than GPT-6 Luna through its Responses API.
  • Three answer types: "predicate" gives the probability that something is true (for example, whether a product shows visible damage); "choice" picks one of the developer's predefined options and adds probabilities and confidence; "score" rates an input against an ordered rubric.
  • What's new since the public beta: a native integration with GPT-Live, OpenAI's voice model; caching for reused context; processing in the US and the EU; and up to 400 questions per request, twice the previous limit. OpenAI also says it added capacity in several regions in Asia.
  • Pricing: $0.10 per million input tokens and $0.01 per million cached input tokens. OpenAI says you only pay for input tokens: there is no output charge. Zero Data Retention is available to eligible customers.
  • Who uses it: the announcement names Box, Datadog, Ramp, Sierra and Speak. In a quote from Box in that announcement, the company says that in its own evaluation the API picked the next step for 30 certificate-of-insurance cases about 5 times faster, with the same accuracy as the Responses API, and sent lower-confidence cases to reviewers.
  • The competition: according to TechCrunch, the "decision model" category was started by Jev, from TypeSafe, and OpenAI's model ranks behind Jev on two third-party leaderboards. Microsoft released its own model of this kind, Microsoft-Decision-1, on October 9 at $0.042 per million input tokens with free output, according to Microsoft.

What changes and what doesn't

What changes: OpenAI's decision model is no longer a beta. It is open to any API customer, with published pricing and the option to process data in the US or Europe.

What doesn't change: it is a developer tool, not a ChatGPT feature or something you switch on from your website's admin panel. And it doesn't write: if you need a drafted reply, you still need a chat model.

How to tell whether it fits your website

  1. Find the decisions you make over and over. Contact-form messages that go to sales, support or spam; reviews or comments that need approval; chatbot conversations that should go to a person. These are questions with only a few possible answers, which is what the API is built for.
  2. Do the math. At the published rate, 10,000 messages of roughly 500 tokens each, instructions included, add up to 5 million input tokens: about $0.50. That is this article's estimate; real usage depends on how long your instructions are and how much caching helps.
  3. Test before you automate. Take around a hundred real cases you already know how to classify and compare the API's answers with yours. OpenAI offers a playground for this.
  4. Decide what happens when it's unsure. Because the API returns a probability or confidence, set a threshold: above it, the decision applies automatically; below it, a person reviews it. That is what Box describes in its example.
  5. Be careful with personal data. Forms carry names, emails and phone numbers. Before sending them to an outside service, check which fields you actually need to send, where they are processed and whether your privacy notice covers it.

Related: Microsoft-Decision-1: Microsoft's AI That Only Decides

Sources

Updates: if OpenAI changes the Decisions API's pricing or limits, it will be added here with a link.