OpenAI makes its Decisions API generally available for classifying, routing and scoring
The endpoint returns a probability, a choice or a score instead of generated text, runs only on gpt-6-luna, and charges for input tokens but not output tokens.
OpenAI made its Decisions API generally available on October 22. The changelog entry for that day reads: "The Decisions API is now generally available with gpt-6-luna." It had been in public beta since October 6. Instead of writing text, the endpoint answers questions about an input with typed results. OpenAI's guide says it "evaluates text, images, and JSON and returns typed answers about 10x faster than the Responses API."
What it returns
Each request carries shared input and a list of questions, and each question has one of three types. A predicate returns "an estimate from 0 to 1 that the condition is true." A choice picks "one of your supplied values." A score rates the input against ordered levels and returns "the probability-weighted average of the level indices." General availability raised the limit to "up to 400 questions, increased from 200 in public beta," and added prompt caching, images inside question instructions and JSON input. The guide says gpt-6-luna is "the only currently supported model."
On price, the guide states that "uncached input costs $0.10 per 1M tokens" and "Output tokens are not charged," with regional processing premiums and long-context multipliers still applying.
Why it matters
A lot of AI spend goes on calls that only need a label: is this ticket urgent, which team gets this email, does this photo show damage. Paying a general model to write out that label in text is the expensive way to get it. OpenAI's own advice is to set thresholds "based on the cost of false positives and false negatives," which means testing against examples where you already know the right answer before switching anything in production.