OpenAI makes the Decisions API generally available for low-latency UI branching
OpenAI made the Decisions API generally available on October 22. Powered by GPT-6 Luna, it returns a probability, choice or score from text, images or structured input, and can drive live UI state such as avatar expressions through GPT-Live decision.event messages.
Why it matters
Generative interfaces need more than component generation: voice and interaction hot paths must quickly choose whether to clarify, escalate, retry or react. Bounded answers separate those branches from long-form generation and create a narrower, more testable runtime contract.
A closer look
Developers define a question and allowed answers, then choose predicate, choice or score. A request can contain up to 400 questions, with native JSON, images, reused-context caching and US/EU processing residency. Native GPT-Live integration can observe transcript deltas and emit decision.event; the official example maps a selected reaction to avatar animation. OpenAI says the dedicated endpoint can be up to 10× faster than GPT-6 Luna through the Responses API and lists early self-evaluations from Box, Datadog, Ramp, Sierra and Speak.
Who this is for
Product and platform teams building low-latency voice interfaces, agent routing, live reactions and constrained UI decision layers
Original sources
Revisions & corrections (1)
— Backfilled an official October 22 product event with direct generative-interface relevance: bounded runtime decisions and GPT-Live UI reactions. Date, availability, answer types and published examples were verified in the primary announcement; vendor metrics remain attributed and caveated.