On October 6, 2026, OpenAI introduced the Decisions API — a tool designed to transform text and visual inputs into concrete actions within applications.
On Wednesday, October 6, 2026, OpenAI published a video on its YouTube channel promoting a new product — Decisions API. This is not just another language model or an update to an existing tool. It is an API designed to quickly and precisely turn text, images, or other inputs into specific actions within an application. It aims to answer the question: "What's next?". Because the goal here isn't for AI to suggest — it's to decide.
How does the Decisions API work?
The scheme is simple yet elegant. The developer provides the model with three elements:
- Input — text, image, or another data format,
- Question — a specific query intended to lead to a choice,
- Response options — a limited list of possible actions.
Based on this data, the model selects one of the options. And that's it. No long analyses, no interpretations. Choice = action.
This process is powered by GPT-6 Luna — a new OpenAI model which, as highlighted in the video, has been optimized for rapid decision-making within a limited number of choices. This is not a general-purpose model. Its power lies in its precision and response time.
Why can few options mean a lot?
Many AI algorithms work by searching for the best answer among hundreds of possibilities. That is powerful, but slow. The Decisions API takes the opposite approach — limiting the scope to a few key choices is intended to make the system more responsive. This isn't theory. It's design.
What was shown in the video?
The video is not limited to technical descriptions. It presents four specific use cases that demonstrate how the Decisions API can impact everyday applications:
1. Text classification and lead routing
The system receives a lead description, a question like "Which category does this belong to?", and three options: A, B, C. Based on the model's choice, the lead is automatically routed to the appropriate sales team. No manual intervention required.
2. Decisions based on images
The video demonstrates robotics that analyze a photo of an object and decide how to handle it — e.g., "move to shelf 3" or "block". This isn't recognition — it's action.
3. Real-time voice expression
The Decisions API works with the GPT-Live-1 model, which generates voice responses based on the context of the conversation. The model decides which emotion — joy, discouragement, curiosity — should be conveyed in the response. This isn't translation. It's reaction.
4. Robotics — Lavender control
One of the most interesting parts of the video is the demonstration of the "Lavender" robot, whose movement is decided based on situational analysis. This isn't hard-coded programming. It's adaptation.
Where does the Decisions API fit into the OpenAI ecosystem?
The video also features two other models: GPT-Live-1 and Microduck. Their roles are not fully explained, but the context suggests they are specialized models — one for voice, the other for vision. The Decisions API acts as a bridge between them. It is not a monolith. It is an orchestra.
This is a step toward what could be called modular AI — where many models each perform their part well, and their collaboration creates a complete response.
What we still don't know
The video leaves several key questions unanswered:
- What are the exact hardware requirements for GPT-6 Luna?
- Will the Decisions API be available to all developers, or only to selected partners?
- What are the operational costs?
- What are the limitations on the number of options or input types?
None of this is in the video. This is not a lack of information — it is by design. OpenAI is presenting a concept, not documentation.
What does this mean for developers?
If you have an application that needs to react quickly to inputs — such as a customer service system, robotics, or a voice interface — the Decisions API might be what saves you from having to build your own decision-making models from scratch.
However, keep in mind: this is not a tool for everyone. If your application requires a wide range of possible answers, this is not for you. The Decisions API only works when the number of choices is small.
FAQ
Is the Decisions API available today?
It is unknown. The video does not specify whether this is an announcement, a beta, or a production version. We have to wait for an official announcement.
How is this different from the ChatGPT API?
The ChatGPT API is a language model — it generates text. The Decisions API is an interpreter — it chooses an action. This is not an upgrade. It is a different product.
Is GPT-6 Luna a new model?
Yes. In previous posts, I wrote about GPT-5.6 Luna — now it's time for GPT-6 Luna. This is not an iteration. It's a breakthrough.
No. This marks the beginning of decision engineering. Instead of asking "what will I say?", you ask "what will we do?". It's a different language.
How does this relate to GPT-6 Astra?
Astra was the last OpenAI model aimed at developers. The Decisions API is not a model — it is a layer on top of models. It can be their partner.
The video "Introducing the Decisions API" is not just a presentation. It is an invitation — to rethink how AI should work: not as an assistant, but as a decision-maker.
Check it out for yourself. Because you can't evaluate what hasn't been launched yet.
Comments