OpenAI makes the Decisions API generally available

OpenAI has moved its Decisions API out of beta and into general availability. The API is built on a model called GPT-6 Luna and is designed to answer narrow, bounded questions about text, images, or structured JSON input rather than generate open-ended text. A single request returns one of three answer types: a probability that a condition is true, a choice from a fixed set of options, or a score against a rubric.

Since the public beta, OpenAI says it has added prompt caching for reused input, the ability to attach images to individual questions, native JSON input, support for up to 400 questions per request (up from 200), hosted image URL inputs, conversation history evaluation, organization-level usage reporting, and direct integration with OpenAI's GPT-Live voice product.

OpenAI says the Decisions API answers questions roughly 10 times faster than routing the same questions through its Responses API, and cites customer results including Box, Datadog, Ramp, Sierra, and Speak reporting faster or cheaper classification and routing tasks. Those performance and cost figures come from OpenAI's own announcement.

Why it matters

A cheaper, faster endpoint for yes/no and classification-style calls could replace a lot of general-purpose LLM calls currently used just to route or score requests.

Source: OpenAI