Decisions API: classification, routing, and scoring
Returns bounded answers with GPT-6 Luna; up to 400 questions per request.
- Input/output: Provide text, images, or JSON plus a question. Receive a predefined choice, probability, or rubric-based score.
- Use: Route documents or support requests; choose whether an agent continues, retries, or asks for clarification.
- Release: General availability with GPT-6 Luna, prompt caching, native GPT-Live integration, and up to 400 questions per request.
- Price at launch: $0.10 per million input tokens; $0.01 for cached input. No output-token charge. Cache writes have separate pricing.
- Implementation requirement: Define the allowed answers, evaluate labeled examples, and decide what uncertainty should send to a reviewer. The endpoint supplies a decision; your application executes the action.
OpenAI announcement · October 22, 2026 · General availability.