TAU-HOME.COM
LOADING

OpenAI Releases Decisions API in Public Beta, Powered by GPT-6 Luna

OpenAI launched the Decisions API in public beta powered by GPT-6 Luna, offering fast predicate, choice, and score evaluations at $0.10 per 1M input tokens.

tau · October 9, 2026

#OpenAI #Decisions-API #GPT-6-Luna #AI-Agent #API

OpenAI Releases Decisions API in Public Beta, Powered by GPT-6 Luna

On October 6, 2026, OpenAI officially rolled out the Decisions API (POST /v1/decisions) in public beta, introducing a specialized endpoint designed to evaluate structured judgments and decisions over text and image inputs. Powered by the gpt-6-luna model, the API offers response latencies approximately 10 times faster than the conventional Responses API, targeting agentic control flow, routing, and real-time decision pipelines.

Diagram of OpenAI Decisions API architecture and decision-tree routing powered by GPT-6 Luna

Image source: Unite.AI / OpenAI

Traditional text generation endpoints—even when configured with JSON mode or Structured Outputs—suffer from latency overhead and decoding costs because they sequentially generate full output tokens. OpenAI designed the Decisions API to serve as a lightweight, low-latency evaluation layer tailored specifically for agent branching and deterministic classification tasks.

Three Decision Types: predicate, choice, and score

The Decisions API provides three core evaluation primitives to answer developer-defined questions over provided context:

  • predicate: Evaluates the probability of a specific condition being true or false. Ideal for guardrail checks, conditional gating, and fast filtering.
  • choice: Classifies input across a predefined list of discrete options. Useful for intent recognition and task routing.
  • score: Computes a probability-weighted average score across an ordered level schema. Designed for priority evaluation and quantitative quality scoring.

Developers can pack multiple independent questions into a single request array, evaluating several criteria in a single round trip to minimize network latency.

Pricing and Input Specifications

The Decisions API introduces a streamlined cost structure and input protocol:

  • Pricing: Running on the gpt-6-luna model, requests are billed at a flat rate of $0.10 per 1M input tokens. There are no fees for output tokens or cache read/write operations.
  • Input Specifications: The API accepts both text and image inputs. Currently, image inputs must be supplied as inline base64 data URLs; hosted HTTP/HTTPS URLs and file_id references are not supported.

Integration Caveats and Constraints

When integrating the Decisions API into production systems, developers should account for the following architectural constraints:

  • Supported Models: Currently, evaluations are supported exclusively on the gpt-6-luna model.
  • Sequential Dependencies: Questions batched within a single request array are evaluated independently. Workflows requiring dependent follow-up evaluations contingent on earlier outputs must be handled via sequential branching calls rather than a single batched payload.

Sources