OpenAI announced on October 22, 2026 the general availability of the Decisions API, a dedicated endpoint for structured, low-latency responses. Built on GPT-6 Luna, it focuses the model’s intelligence on questions with bounded answers, enabling classifications, routing and scoring up to 10 times faster than the same model through the Responses API.
What changed in GA
Since public beta (early October), OpenAI added native GPT-Live integration, prompt caching, inference residency in the US and EU, native JSON inputs, and the ability to include images in individual question instructions. The limit rose to 400 questions per request (double the beta). Latency was also improved, with additional capacity in several Asian regions.
Pricing remains $0.10 per million input tokens ($0.01 for cached input). There is no charge for output tokens. Zero Data Retention and HIPAA eligibility are available to qualified customers.
Response types
- Predicate: estimates the probability (0 to 1) that a condition is true.
- Choice: selects one predefined option, with confidence.
- Score: assigns a probability-weighted score against ordered levels.
Why it matters
Agents and applications make dozens or hundreds of micro-decisions in every workflow (classify, route, choose next action). The Decisions API lowers the cost and latency of these operations, making it practical to place them on the hot path of production systems. Companies including Box, Datadog, Ramp, Sierra and Speak have reported speed gains (3x to 40x) and cost reductions without accuracy loss.
Native GPT-Live integration enables more responsive voice experiences, such as avatars that react in real time or support agents that quickly decide whether to clarify, escalate or complete an action.
Sources
Transparency: This content was created, edited or reviewed with the assistance of artificial intelligence. Information was cross-checked with public X posts and available internet sources. Consult original sources for full context.
By GeekikiBot
