OpenAI Decisions API Hits Public Beta With 10x Faster Typed Answers

OpenAI has put its Decisions API into public beta, and the pitch is pretty specific: feed it text, images, or both, and get back typed answers your code can act on. No prose. No post-processing a blob of model text into a label. Just probabilities, choices, and scores, wrapped in an API call that OpenAI says is about 10x faster than its Responses API.

That matters because a lot of production AI work still looks surprisingly clunky. Ask a model a question, then parse the answer, then hope the formatting holds. Decisions is aimed at that exact pattern. The request format is simple too: model, input, and questions. Each question gets a name, a type, and instructions, and the response comes back as an answers array keyed to those names.

OpenAI is only offering one model here right now, gpt-6-luna, through its hosted API. The endpoint is POST /v1/decisions, with a Playground available as well. The company says general availability should arrive in the coming weeks. The model card doesn’t disclose parameter count, but it does list a 1,050,000-token context window.

The three question types are built for different jobs. Predicate checks whether something is true and returns a probability between 0 and 1. Choice picks one option and also returns per-option probabilities plus a confidence field. Score assigns a value to ordered levels, using a probability-weighted average of level indices. OpenAI’s own example puts probabilities of 0.1, 0.7, and 0.2 at a score of 1.1, between “Workaround available” and “Fully blocked.”

The company is drawing a line around where this sits in its stack. Use Decisions for probabilities, choices, or scores. Use Structured Outputs when you want JSON to fit your own schema or need explanations. Use function calling when a model needs to ask for a tool with arguments. OpenAI also says it has not published accuracy or calibration data for the endpoint, and recommends setting thresholds with labeled examples from your own application. Pricing starts at $0.10 per 1M input tokens, with no output, cache-read, or cache-write charges, though regional premiums and long-context multipliers still apply.