OpenAI Launches Decisions API in Public Beta for Fast Content Classification

OpenAI has released the Decisions API in public beta, a new endpoint designed for applications that need fast, structured evaluations of text, images, or both. According to the company's documentation, the API returns typed answers about 10x faster than the Responses API, making it suitable for classifying content, routing requests, and prioritizing work.

How the Decisions API Works

The API is accessed through a dedicated POST /v1/decisions endpoint. Each request has three parts: the model, the input, and the questions. The input is shared evidence for the questions and can be a text string or user messages containing text and images. The questions define what to evaluate, including each question's type, instructions, and any allowed choices or score levels. The response contains an answers array, and each question should be given a unique name that the API echoes back in the response.

Question Types and Use Cases

Developers can choose from three question types. A predicate question checks a condition, such as visible damage or passage relevance, and returns a probability from 0 to 1 that the condition is true. A choice question selects one option from a fixed set, such as a department or content category, and returns the chosen value. A score question rates an input against ordered levels, such as issue severity, and returns the probability-weighted average of the level indices, which can fall between levels.

OpenAI recommends using choice for categories without an order, such as departments, and score for ordered levels such as severity. The documentation notes that Structured Outputs with the Responses API remains the right tool when an application needs to generate an object following a custom JSON schema, and function calling is appropriate when a model needs to request a tool call with arguments.

Availability and SDK Support

The Decisions API is currently in public beta, and OpenAI expects it to reach general availability in the coming weeks. The gpt-6-luna model is the only model currently supported. SDK examples are available for Python 3.26.0, JavaScript 7.30.0, Go 3.73.0, Ruby 0.101.0, and Java 4.78.0 or later. Images must be provided as inline base64 data URLs; hosted HTTP or HTTPS image URLs and file_id inputs are not supported by this endpoint. Developers can experiment with questions and inputs in the Playground before writing code.

Sources: developers.openai.com