OpenAI's Decisions API runs on gpt-6-luna alone, and the model's own endpoint table omits it
OpenAI released a Decisions API in beta on October 6, an endpoint that hands back a probability, a fixed choice or a numeric score instead of prose, and gpt-6-luna is the only model it accepts. The route is its own, POST /v1/decisions, and OpenAI's API changelog says it turns text and images into typed answers "10x faster than the Responses API", a comparison the company publishes with no benchmark attached.

OpenAI documents the Decisions API as a public beta on gpt-6-luna, with general availability given only as the coming weeks. Image: OpenAI
The model it runs on does not list it. The gpt-6-luna model page carries an endpoint support table of 18 rows, counted row by row at 22:40 UTC on October 6, and v1/decisions is not among them. Chat Completions, Responses and Batch are the three marked Supported, and the other 15 rows are marked Not supported.
Three question types, and one that lands between its own levels
A request carries three fields, model, input and questions, and every question picks one of three types. A predicate returns a probability from 0 to 1 that a condition holds, and a choice returns one of the values the developer supplied, with a probability for each and a separate confidence figure.
The third type is the odd one. A score returns the probability-weighted average of ordered level indices, so OpenAI's own worked example turns probabilities of 0.1, 0.7 and 0.2 across three severity levels into a score of 1.1, a number that corresponds to none of them. Levels are indexed from 0, and OpenAI writes that the score "can fall between levels".
Each question takes a name that the API echoes back in its answers array, and any question can return a refusal rather than a scored answer. Questions in one array share the same input, and OpenAI says a decision that depends on an earlier answer needs a second request.
Inline images only, and no charge for output
Images have to arrive as base64 data URLs inside the request. OpenAI states that hosted HTTP or HTTPS image URLs and file_id inputs "aren't supported by this endpoint", caps one request at 128 image parts, and rules out non-user roles, function calls, function-call outputs, files, audio and item references.
On price, OpenAI's guide puts input at $0.10 per 1M tokens and says there are "no cache-read, cache-write, or output-token charges", while regional processing premiums and long-context multipliers still apply. That is a different shape from gpt-6-luna's published rates elsewhere, which list $0.1 per 1M input tokens, $0.01 cached input, $0.125 cache writes and $0.5 output, and it follows the price cut OpenAI applied to GPT-6 Sol and Luna in September.
Still a beta, with US and EU residency
The endpoint supports Zero Data Retention and HIPAA use for eligible customers, and data residency and regional processing in the United States and in Europe, which OpenAI defines as the EEA plus Switzerland. There is a playground for it on the OpenAI platform, at platform.openai.com/decisions.
OpenAI calls the beta public and says it expects general availability "in the coming weeks", which is not a date. The classification work the endpoint is built for has until now gone through the Responses API, the same route behind the hosted browser OpenAI added to its Agents API, and for the moment Decisions has one model behind it and no entry in that model's support table.