OpenAI Launched Decisions API for Faster Inference
The new API provides structured outputs with 150-millisecond latency for real-time decision-making applications.
Updated on Sept. 29, 2026 in Artificial Intelligence
Live Poll
Do you trust AI tools that provide explicit confidence scores with their answers?
OpenAI has announced and released a limited preview of the Decisions API, a specialized tool powered by the Luna model that provides predefined answers and confidence scores. The API is designed to classify content or select actions from limited sets of choices.
Why it matters
The API aims to provide low-latency inference capabilities for developers needing quick, constrained decision-making within their applications. It fills a performance gap for workflows requiring immediate outcomes compared to standard, generative model response times.
The Decisions API delivers a response time of 150 milliseconds, significantly faster than the 1.6-second latency observed in the GPT-6 Luna model. It functions by ingesting developer-provided questions and context, then outputting a classified result with an associated confidence score.
The players
OpenAI
An artificial intelligence research and deployment company known for its large language model stack and development of the Luna and GPT models.
The details
The Decisions API utilizes the Luna model to categorize input content or execute specific actions from a constrained list of options. By restricting the model to a defined set of outcomes, the system avoids the overhead of open-ended text generation. This architectural constraint allows the API to achieve rapid response times suitable for real-time automated workflows.
Timeline
September 29, 2026: OpenAI announced the Decisions API at its annual DevDay conference.
Coming days: A broad release of the Decisions API is expected.
The Tech Race
The Decisions API marks a strategic pivot toward constrained, high-speed inference to compete in the real-time automation market. It complements the broader capabilities of the GPT-6 Luna model by sacrificing generative flexibility for raw response speed.
Developers building automation or classification tools will gain access to a faster inference path once the broad release launches in the coming days. The API is currently restricted to a limited preview, requiring early adopters to wait for the next phase of deployment.
The takeaway
The Decisions API represents a shift toward optimizing models for specific, low-latency decision tasks rather than general-purpose generation. Watch for the broad release announcement in the coming days to determine the API's production feasibility for scaling high-frequency enterprise workflows.
What happens next
OpenAI plans a broad release of the Decisions API in the coming days, which will allow for wider integration beyond the current limited preview.
Further reading
For more developments in high-speed model architecture, explore the latest research in /Artificial Intelligence.
Source note: This article includes information reported by The New Stack.
Live Poll
Do you trust AI tools that provide explicit confidence scores with their answers?
