OpenAI made the Decisions API generally available on October 22, 2026. Built on GPT-6 Luna, it focuses the model on “questions with bounded answers,” letting applications and agents make near real-time decisions from text, images and structured inputs “up to 10x faster than GPT-6 Luna through the Responses API.”
Developers define a question and its allowed answers and supply context. The API returns one of three answer types: a predicate (a probability from 0 to 1 that a condition is true), a choice (one of the supplied values, with probabilities and a confidence), or a score against ordered rubric levels. Since the public beta OpenAI has added a native GPT-Live integration, prompt caching, US and EU inference residency, native JSON inputs, images in question instructions, and up to 400 questions per request, twice the previous limit.
Pricing is $0.10 per million input tokens and $0.01 per million cached input tokens, and “You only pay for input tokens: there are no output-token charges.” Customer quotes on the page include Datadog reporting 40 times faster p95 latency than other frontier models with comparable accuracy for scoring the risk of agent-generated shell commands, Sierra reporting under 280 ms p95 and 90 percent lower cost per call, and Speak reporting time-to-next-action falling from 1.4 s to 0.13 s inside GPT-Live.
Why it matters: agents make many small routing and gating decisions, and a dedicated, input-priced endpoint for them turns those calls into a separate, cheap product line. It also gives builders a fast place to put safety checks, such as scoring a command before it runs, in the hot path of an agent.
What it does not show: the speed and cost figures are OpenAI’s and its customers’ own and are not independently measured, and the “10x” comparison is against Luna through the Responses API rather than against competing classifiers.