OpenAI Releases Decisions API in Public Beta — 10x Faster Than Responses API

OpenAI has opened its Decisions API to all developers in public beta, a specialized endpoint that returns structured answers — yes/no predicates, ranked choices, or severity scores — at roughly ten times the speed of its existing Responses API. The launch targets a class of production AI applications where latency and cost matter more than verbose reasoning.

Most AI API calls today return freeform text that developers then parse, validate, and route through their own logic. The Decisions API flips that model: the developer describes a decision to make, and the API returns a typed result the application can act on directly. This architecture is common in high-frequency systems — fraud detection, content moderation, clinical triage — where milliseconds and predictable output formats are non-negotiable.

Technical specifics: The Decisions API public beta is powered by GPT-6 Luna, a latency-optimized variant of the GPT-6 family. Pricing is set at $0.10 per million input tokens, with output tokens unbilled — a significant cost advantage for high-volume workloads that generate short outputs. Three return types are available at launch:

• Predicates — boolean true/false decisions (e.g., "Is this message spam?")
• Choices — ranked selection from a developer-defined list (e.g., route to billing, support, or escalation)
• Severity scores — numeric ratings on a defined scale (e.g., content risk 1–10)

The endpoint also ships with HIPAA eligibility and Zero Data Retention (ZDR) support from day one, which removes a common compliance blocker for healthcare and financial services builders. General availability is expected within weeks, according to OpenAI.

For developers and enterprises, the Decisions API collapses a multi-step pattern — prompt → parse → validate → route — into a single typed call. Teams building clinical triage tools, e-commerce recommendation engines, or compliance screening pipelines no longer need to maintain brittle output parsers that break when the model's phrasing changes. The zero-output-billing model also makes it economical for pipelines that run thousands of short decisions per minute.

The HIPAA and ZDR support at launch signals that OpenAI is prioritizing regulated industries as early adopters, rather than positioning this as a developer toy.

What to watch: Anthropic's tool-use and structured output capabilities and Google's Gemini function calling serve similar use cases. As OpenAI moves toward typed, task-specific endpoints, watch whether the other major providers respond with equivalent latency-optimized structured output APIs — and whether OpenAI bundles Decisions API access into its enterprise agreements to accelerate adoption.