What is OpenAI's Decisions API, the model that never writes a sentence?
Most of what businesses do with AI is not conversation. It is a million tiny decisions. Is this email spam? Which team should get this ticket? Which of these three tools should the agent call next? You do not need an essay for that. You need an answer, and you need it fast and cheap.
That is the bet behind OpenAI's new Decisions API, built on GPT-6 Luna. It was announced at DevDay on 29 September and opened to all developers in early October, with general availability expected within weeks.
What you get back
You give it an input and a question with a fixed set of possible answers. It does not write anything. It hands back one of three typed results: a predicate (how likely a statement is to be true), a choice (one option from your list, with a confidence score), or a score (a probability weighted rating on an ordered scale). Text and images both work.
- Speed: up to 10 times faster than calling the same model through the regular API, with answers in around 150 milliseconds.
- Price: $0.10 per million input tokens, nothing for output tokens, and nothing for cache reads and writes.
- Jobs it is pitched for: classifying content, routing requests, and choosing the next action for an agent.
And here is what looks odd: why would OpenAI charge nothing for output and caching?
Part of the answer is that the output is tiny by design, so there is almost nothing to bill. One plausible read of the rest: the simpler the invoice, the easier it is to win over anyone who counts every cent. That is a reading of the price list, not something the company has said.
The timing is not subtle either. It arrived weeks after a small startup shipped a model built for the very same job.