Source · docs
OpenAI Decisions API guide
Summary
OpenAI released the Decisions API in public beta on 2026-10-06 (per the API changelog).
It is a separate endpoint, POST /v1/decisions, that returns typed answers to questions
about text or images instead of generating free text. OpenAI says it is about 10x faster
than the Responses API. It runs only on gpt-6-luna.
Key points
- A request has a model, an input (text, or a user message with text and inline base64 images), and a list of questions.
- Three question types: predicate (probability 0–1), choice (one value from a fixed set plus a confidence), and score (probability-weighted average over ordered levels).
- Intended uses: classification, routing, prioritisation, and picking an agent's next action.
- Price: $0.10 per 1M input tokens, with no output or cache charges (as of 2026-10-07). Regional processing premiums and long-context multipliers still apply.
- Hosted image URLs and file IDs are not supported as input.
- Supports Zero Data Retention and HIPAA eligibility; available in the US and Europe (EEA and Switzerland). OpenAI expects general availability "in the coming weeks".
- Secondary coverage dates the announcement to DevDay on 2026-09-29 with ~150 ms latency claims; the changelog dates the beta release to 2026-10-06.
Why it matters
Classifier-style calls (routing, moderation, triage) have usually been done with full generation plus structured outputs. A dedicated endpoint that returns probabilities and charges only for input makes these calls much cheaper and faster.