OpenAI's new API makes real-time calls using GPT-6 Luna

Writing complex prompts just to get an LLM to pick between option A and option B has always felt like overkill.

OpenAI just launched the Decisions API, a new tool powered by GPT-6 Luna designed to give applications real-time decision-making skills without the usual prompt engineering headaches.

Multiple-choice routing for AI agents

The setup is simple: developers define a question and a list of possible answers.

The API accepts text or images as context. For instance, you can pass in a customer support ticket alongside a list of internal teams, and the API will return the right department to handle it. Beyond request routing, it is built for content classification and determining an agent’s next logical move.

It is currently rolling out in a limited preview to select API users, with broader availability planned within days.

Why it matters

Developers spend way too much time wrestling with structured outputs and system prompts just to route user queries. Decoupling basic logic decisions into a dedicated, Luna-powered API should make agents significantly faster and cheaper to run.

We think this is a smart move by OpenAI to capture the plumbing of agentic workflows. Instead of burning heavy compute on simple branch logic, Luna handles the switchboard.

The catch? Handing over your application's core control flow to a probabilistic model always comes with edge-case risks. If you decide to adopt it, you will still want solid deterministic guardrails in place for when it picks the wrong bucket.