OpenAI has opened its Decisions API to everyone with API access. General availability arrived on October 22, 2026, with pricing set at $0.10 per million input tokens.
Output costs nothing. Neither do cache reads or cache writes.
What the Decisions API actually does
The Decisions API is designed for high-volume, low-latency decision-making. Instead of returning paragraphs, it produces structured, typed answers in the form of probabilities, choices, and scores.
The endpoint runs on OpenAI’s GPT-6 Luna model. It targets routing requests, classifying inputs, and repeated micro-decisions.
One request can carry up to 400 questions. That ceiling was raised from what developers had during the beta period.
Speed, features, and compliance
The Decisions API is reported to run up to 10 times faster than using GPT-6 Luna through the standard Responses API for bounded decisions.
In demonstrations, latency came in at approximately 150 milliseconds.
OpenAI also shipped several upgrades between beta and general availability. These include native GPT-Live integration, prompt caching, and support for image and JSON inputs.
On the compliance side, the API supports Zero Data Retention and is HIPAA eligible. It is positioned to serve customers in both the US and European markets.
A fast path from preview to launch
OpenAI first showed the Decisions API in limited preview at DevDay on September 29, 2026. A public beta followed on October 6. General availability landed on October 22, a little over three weeks after the first preview.
The jump in per-request question capacity and the added input types also show the beta period produced concrete changes.
What this means for developers and the AI market
By charging only for input, OpenAI makes the cost of any given call predictable from what a developer sends in. Output of a classification is typically a single choice or a score, so the meaningful cost driver is the context being evaluated.
The 10x speed figure applies to bounded decisions and comes framed as an upper bound, and the 150 millisecond latency was measured in demonstrations.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.