OpenAI Releases Decisions API in Public Beta for Faster Content Evaluation

How Developers Can Use Typed Responses in Practice

OpenAI has launched the Decisions API in public beta, offering developers a tool to evaluate text, images, or both and receive typed responses up to ten times faster than the Responses API. The service enables applications to determine the likelihood of a condition being true, select from predefined options, or score inputs against custom rubrics. This release targets developers seeking to automate content classification, request routing, and workflow prioritization within their software systems.

The Decisions API builds on OpenAI’s existing model infrastructure but optimizes for speed and structured output in decision-making tasks. By returning probabilities, discrete choices, or rubric-based scores, it supports use cases such as filtering inappropriate content, directing user inquiries to appropriate support channels, or ranking items for review. Early access participants can test the API through OpenAI’s developer platform, with documentation guiding integration into applications requiring real-time, reliable assessments.

What Advantages Does Speed Offer in Real-Time Systems?

The API allows developers to define specific conditions—for example, whether an image contains violence or if a text snippet expresses intent to purchase—and receive a probability score indicating confidence. Alternatively, it can return a choice from a fixed list, such as categorizing a support ticket as „billing,” „technical,” or „general.” For more nuanced evaluation, users can apply a custom rubric to score content on dimensions like clarity, relevance, or safety. These outputs are designed to be consumed directly by application logic, reducing the need for post-processing.

A tenfold improvement in response time means the Decisions API can handle high-volume scenarios where latency impacts user experience, such as live chat moderation or dynamic content feeds. Faster evaluations allow systems to make split-second decisions without noticeable delay, improving responsiveness in interactive applications. OpenAI emphasizes that this performance gain does not come at the cost of accuracy, as the API leverages the same underlying models tuned for efficiency in decision-specific tasks.

What types of inputs can the Decisions API process? The API accepts text, images, or a combination of both as input, enabling multimodal evaluation for applications that require analysis across different media formats.

Frequently Asked Questions

How is the Decisions API different from the Responses API? While both APIs use OpenAI’s models, the Decisions API is optimized for returning structured, typed outputs like probabilities, choices, or scores, and delivers them approximately ten times faster than the Responses API, which is designed for more open-ended, generative tasks.

Can developers customize the rubrics used for scoring? Yes, developers can define their own scoring criteria based on application-specific needs, allowing the API to evaluate content against tailored dimensions such as toxicity, relevance, or completeness.