OpenAI makes Decisions API generally available for faster automated actions

OpenAI has released its Decisions API out of beta into general availability, introducing a dedicated way for applications to make fast judgements without generating long text. Instead of asking a model to produce conversational prose and then parsing the output, the endpoint evaluates text, images or structured data to answer specific, preconfigured questions.

The tool runs on OpenAI's GPT-6 Luna model and is engineered for tasks where the set of possible answers is already known. OpenAI states that focusing the model purely on selection rather than sentence generation lets the API return results up to ten times faster than standard model calls.

How structured judgements work

The system handles three distinct types of questions:

• Predicate checks calculate the probability that a specific statement is true, such as verifying whether a submitted photo shows accidental damage.
• Choice questions select the single best option from a predefined list and report a confidence score, which software can use to triage support tickets or dispatch field technicians.
• Scoring questions evaluate incoming content against a numerical rubric, producing a weighted rating for quality assurance or lead qualification.

With the full release, developers can include up to 400 questions within a single API request, double the limit available during the trial. The company also added direct support for JSON data inputs, prompt caching to lower computing expenses on repeated reference material, and live voice integration.

Pricing for the endpoint sits at $0.10 per million input tokens, dropping to $0.01 per million tokens when content is retrieved from cache. OpenAI does not bill for output tokens on the endpoint. Commercial software companies including Box, Datadog, Ramp and Sierra have already started using the tool in production.

What this means for Australian businesses

Many Australian small and medium businesses have struggled with the delay and expense of using conversational AI for basic administrative workflows. Waiting several seconds for an AI to draft a full explanation when all an application needs is a simple yes, no or category tag slows down customer forms and raises operational bills.

Because this tool bypasses text drafting and charges only for incoming tokens, Australian companies can automate repetitive back-office decisions much more cheaply. Practical uses include instantly routing customer inquiries to the right department, scoring incoming sales leads from web forms, and checking insurance or trade claim photos before staff open the file.

The structured format also reduces integration bugs. By constraining outputs to strict options or numerical probabilities, business software avoids the formatting errors and unexpected answers that often cause traditional chatbot automations to fail.