OpenAI has introduced a Decisions API for tasks where an evaluation can be expressed as a probability, a choice from set categories or a rating on a scale. It can process text, images or both, and OpenAI says it runs about ten times faster than the Responses API.
The launch also includes a simpler structure for paid API access. Together, the updates address how developers can request evaluations and how organizations move between paid tiers.
Three ways to express an evaluation
The Decisions API supports three response types: yes/no probabilities, picks from predefined categories and scale-based ratings. That lets a developer frame an evaluation around the kind of answer a task needs, rather than asking for an open-ended response.
For example, a yes/no probability can represent whether a photo shows damage. A category pick can route a customer inquiry or classify a document. A scale-based rating provides another format for evaluations that need a position on a scale.
These examples are among the use cases OpenAI identifies. The source does not specify how developers should define categories or rating scales, so those implementation details depend on the task.
Public beta, with limited model support
The API is in public beta, and general availability is coming soon. For now, only gpt-6-luna is supported. The listed input price is $0.10 per million input tokens, and output tokens are free.
Those details give teams an initial picture of access and cost, but the beta status and single supported model are important constraints. The source does not describe other model options or give a date for general availability.
The Decisions API can evaluate text, images or a combination of the two. That makes the same basic response formats relevant to both written material and visual input, including the stated example of detecting damage in photos.
Data handling and intended uses
The API supports zero-data retention and HIPAA-compliant use in the US and Europe. These are notable details for organizations considering evaluations that involve sensitive material. The source does not provide further terms or technical explanations of those options.
OpenAI’s examples point to several practical uses: damage detection in photos, automatic routing of customer inquiries and document classification. Each involves turning some input into a compact result that can guide a next step, such as sorting an inquiry or labeling a document.
The article describes the Decisions API as likely a response to a mini-trend of “decision models” that Jev kicked off in mid-September. That context places the launch alongside interest in models designed to make constrained evaluations.
Paid API access moves to three tiers
Separately, OpenAI has reduced its paid API tiers from five to three: Build, Launch and Grow. Organizations move up automatically when their total credit purchases hit the next threshold.
The monthly usage limits listed for the tiers are $500, $5,000 and $200,000. The announcement gives those limits alongside the new tier names, offering organizations a clearer outline of the paid access structure.
The two updates cover different parts of API use: the Decisions API provides specific formats for evaluations, while the tier changes reorganize paid access. Developers weighing the beta can assess its supported inputs, response types, model availability and price against the evaluation task they want to handle.