OpenAI has introduced the Decisions API, a new tool designed to select from predefined answers and attach a confidence score to each result. Frederic Lardinois reports for The New Stack that the API is currently in limited preview, with a broader rollout planned in the coming days.
The company bases the service on Luna, which it describes as the smallest and least expensive model in its current lineup. OpenAI has not disclosed pricing, limits on the number of possible answers, or whether customers will be able to adapt the model using their own data.
Built for choices, not conversation
Unlike a chat model, the Decisions API is intended to return a selection from a defined set of options. Developers provide the question, the available answers, and relevant context. The model then returns a choice with a confidence score. All of this sounds very similar to Jev by TypeSafe.
That approach could be useful for tasks such as:
- Sorting incoming content into categories
- Sending customer requests to the right workflow
- Choosing the next step for an AI agent
- Flagging requests that need human review
OpenAI says the API can return a result in about 150 milliseconds. According to the company, a comparable request to GPT-6 Luna would take around 1.6 seconds. The speed difference could matter in systems that must process many requests quickly.
Many teams currently ask general-purpose language models to choose from a list in a prompt. They may also use token probabilities as a rough measure of confidence. Another option is a small classifier, which can be fast and inexpensive but requires labeled training data and retraining when categories change.
The Decisions API aims to sit between those options. Its practical value will depend on the missing details, especially price, scale, and the reliability of its confidence scores.
Stay up to date
AI for content creation: the latest tools, tips and trends. Every two weeks in your inbox:
