OpenAI Decisions API enters public beta at $0.10 per million tokens

Teams that classify text or route requests with a chat model and a long prompt have a priced OpenAI product to compare against for the first time. OpenAI has moved its Decisions API from limited preview to public beta, with the price, the regions, and the model all now listed on its documentation page.

The guide says the Decisions API "evaluates text, images, or both and returns typed answers about 10x faster than the Responses API." Input costs $0.10 per million tokens, and the page states "You pay only for input tokens: there are no cache-read, cache-write, or output-token charges." Only one model is available at launch, gpt-6-luna, served from a dedicated POST /v1/decisions endpoint.

What a request looks like

Each call to the endpoint carries three parts: the model name, the input evidence as text or messages that can include images, and a set of questions that define what the model should decide. The response contains an answers array, with each result tied back by a unique question name. Three question types are supported.

A predicate returns a probability for a true or false condition. A choice picks one option from a fixed list. A score rates the input against an ordered set of levels. OpenAI presents those three shapes as the main uses for the API, naming content classification, request routing, and work prioritisation as the examples.

The data and compliance details

OpenAI says the Decisions API supports Zero Data Retention and HIPAA for eligible customers, and that regional processing is available in the United States and Europe (EEA plus Switzerland). The guide calls the current release a public beta and says the company expects general availability "in the coming weeks." Model selection is fixed to gpt-6-luna today, with no on-page detail yet on fine-tuning or custom labels beyond the prompt.

How the price compares

The Hacker News thread for the launch, posted twelve hours before this item, has 280 points and 133 comments at time of writing. Several commenters compare the beta against TypeSafe's Jev model, which they put at roughly $0.042 per million tokens, around two to three times cheaper than OpenAI's new price on text-only inputs. One commenter notes that Jev does not accept images, which the Decisions API does, so a step that classifies both text and images cannot be priced against Jev at all. Those are the thread's own numbers, and nobody there published a repeatable benchmark behind them.

Which step to move first

A chat model answering a prompt that lists ten labels will usually cost more and take longer than a single Decisions call that returns one typed answer with a confidence score. For a text-only classification with no privacy requirement, Jev is cheaper by the HN thread's figures. For a classification that includes images, needs Zero Data Retention, or needs HIPAA, OpenAI's price pays for compliance a smaller provider does not sell. The on-page figures from OpenAI are what a team should plan against until independent latency and cost numbers appear.

Source

Primary: Decisions API Overview, OpenAI.

Hacker News discussion: Decisions API is in public beta, 280 points, 133 comments at time of writing.