What if AI made 'decisions' instead of writing long essays? Introducing OpenAI's Decisions API

AI Summary

OpenAI's newly released 'Decisions API' is a new type of tool that allows AI to select the most probable answer from developer-defined options, rather than generating long, verbose text.

Imagine you have to categorize hundreds of customer inquiry emails every day. Until now, if you asked an AI, “Categorize this email as either a return or a simple inquiry and explain in detail,” the AI would have written a long response, adding not just the email content but also the classification result and a polite explanation. But all we really needed was the single word, “Return.”

On October 6, 2026, OpenAI released the new ‘Decisions API’ to solve this inefficiency [Source 14, Source 15, Source 16]. Moving beyond an AI that is simply good at writing, an era has opened where our systems receive the ‘decisions’ they want instantly.

Why is this important?

It is enjoyable when AI answers fluently in daily conversation. However, for developers building software, it is a different story. If AI adds too much explanation, it creates the hassle of having to refine the resulting data, and it slows down processing speed.

The Decisions API has transformed AI from a ‘knowledgeable chatterbox’ into an ‘efficient practitioner.’ Now, instead of verbose explanations, AI selects clear answers within predefined rules [Source 12]. This will bring tremendous efficiency, especially in fields where quick AI judgment is essential, such as customer service automation, data classification, and content filtering [Source 18].

Easy to Understand: AI like a Multiple-Choice Test

Shall we compare the way the Decisions API works to a ‘multiple-choice test’?

If existing AI models were students writing subjective, descriptive essays, the Decisions API is like a student filling out a multiple-choice answer sheet. When a developer throws a question and options like, “Is this email a (Return / Inquiry / Other)?”, the AI only selects and informs you of the answer with the highest probability from those options [Source 9, Source 12].

By doing this, you can skip the process of analyzing complex sentences and filtering out unnecessary words. As a result, processing speed is up to 10 times faster than the existing method (Responses API) [Source 1, Source 15]. Furthermore, it doesn’t just inform you of the answer; it calculates and informs you of the probability that the answer is correct (e.g., ‘98% probability it is a return’), allowing the system to make more sophisticated judgments [Source 1, Source 9].

Current Status

The Decisions API is currently in Public Beta, and developers worldwide can access and test it [Source 16]. It only operates through the ‘GPT-6 Luna’ model and can be used through the dedicated access path (POST /v1/decisions) provided by OpenAI [Source 13, Source 15, Source 16].

The pricing policy is also attractive. Unlike existing complex fee calculations, it only charges for the cost of inputting data ($0.10 per million input tokens), and it does not charge at all for the AI to output or save results [Source 15]. For developers, an environment has been created where large amounts of data can be processed quickly without worrying about costs.

What’s Next?

This announcement shows that AI has begun to settle in earnest as a component of our systems, moving beyond a giant knowledge warehouse. In the future, it will be common for AI to make countless judgments in real-time invisibly inside the apps we build. Even if you don’t talk to AI, your mobile phone will move much smarter and more agilely based on AI’s decisions.

In simple terms, AI is now ready to be a smart helper that silently makes ‘decisions’ behind the system rather than talking to us.

References

Share this article:

Test Your Understanding

Q1. What is the biggest difference between the newly released Decisions API and existing AI models?

It can write longer articles
Instead of writing text, it chooses one from predefined options
Image generation speed is 10 times faster

Instead of generating verbose text, the Decisions API returns one of the options predefined by the developer for problems such as classification or judgment.

Q2. Which model does the Decisions API operate on?

GPT-4o
GPT-5
GPT-6 Luna

The Decisions API is currently only available through the GPT-6 Luna model.

Q3. What is the pricing model for the Decisions API?

Charged based on input tokens only
Charged based on output tokens only
Monthly subscription fee

The Decisions API charges $0.10 per million tokens for input only, with no costs for output or caching.