OpenAI has officially transitioned its Decisions API out of public beta, making the GPT-6 Luna-powered endpoint generally available to developers seeking rapid and structured text and image evaluation. According to reports and official documentation, the service is priced at $0.10 per million input tokens and is engineered to return bounded, typed answers—such as probabilities, category selections, or rubric scores—significantly faster than standard conversational endpoints.
Sending Requests and Structuring Question Types
Developers interact with the endpoint via a dedicated request structure sent to the decisions route utilizing the gpt-6-luna model as the primary backend. Documentation specifies that requests combine shared inputs—including text strings, conversation histories, or custom tool outputs—with up to 400 specific questions per payload. The API categorizes inquiries into three distinct output types to suit different application logic. Predicate questions yield a binary-style probability between zero and one indicating whether a specific condition holds true, such as detecting visual defects in product photographs. Choice questions pick from developer-defined categories while returning individual probabilities and confidence levels. Meanwhile, score questions evaluate inputs against two to ten ordered rubric levels, computing a probability-weighted average index. Furthermore, structured data can be passed directly using inline JSON structures or base64-encoded image URLs, streamlining workflows without requiring complex string conversions.
New Architectural Additions and Regional Expansion
Following its public beta phase, OpenAI integrated several architectural and functional enhancements into the production release. Among the additions is native support for real-time voice integration, allowing applications to process decisions instantly as transcripts arrive. Developers also benefit from prompt caching mechanisms that offer discounted read pricing for shared input prefixes longer than 1,024 tokens. Geographic availability has broadened to include regional inference residency options across the United States and the European Union, alongside expanded processing capacity in Asian data centers. Additional capabilities include support for hosted HTTP and HTTPS image inputs, organization usage tracking endpoints, and Private Link integration for secure enterprise deployments.
Why This Matters: The Operational Shift Toward Bounded AI Decisions
This general availability release represents a strategic shift in how enterprises deploy frontier models for high-frequency automation and classification tasks. Rather than relying on traditional conversational text generation or custom JSON schema parsing via standard chat completions, engineering teams can now offload deterministic routing, content filtering, and risk scoring to a specialized endpoint optimized for speed and cost efficiency. Early industry evaluations highlight substantial performance gains; firms across enterprise software, data infrastructure, and voice orchestration reported dramatically reduced latencies and lower per-call costs across customer support, transaction categorization, and automated workflow routing. By structuring billing around input tokens while omitting output token charges and providing discounted cache reads, OpenAI lowers the economic barrier for running continuous automated evaluations at scale. For software developers and enterprise platform architects, this means large language models are increasingly viable not just as open-ended chatbots, but as high-speed, reliable components within automated operational pipelines where predictable, bounded outputs are critical for real-time decision-making.
Cover: HaeB · CC BY-SA 4.0 · Wikimedia Commons. Layout and headline text by Signal Titans.
Source: OpenAI Makes Decisions API Generally Available with GPT-6 Luna · Decisions | OpenAI API
SIGNAL TITANS
Leave a Reply