--- url: "https://www.xcademia.com/news/openai-launches-decisions-api-for-faster-lower-cost-application-decisions" title: "OpenAI Launches Decisions API for Faster, Lower-Cost Application Decisions" description: "OpenAI's Decisions API is generally available with GPT-6 Luna, structured outputs, GPT-Live integration, prompt caching, and input pricing details." publishedAt: "2026-10-23T11:50:23.823+00:00" updatedAt: "2026-10-23T12:44:21.492805+00:00" type: news category: "ai-ml" source_name: OpenAI source_url: "https://openai.com/index/decisions-api/" tags: - "#OpenAI" - "#DecisionsAPI" - "#ArtificialIntelligence" - "#GPT6Luna" - "#GPTLive" - "#APIDevelopment" - "#AIApplications" - "#DeveloperTools" --- # OpenAI Launches Decisions API for Faster, Lower-Cost Application Decisions > OpenAI has made its Decisions API generally available, giving developers a faster way to classify inputs, score information, and select predefined actions. Built with GPT-6 Luna, it supports text, images, structured JSON, and real-time voice workflows through GPT-Live. Source: **OpenAI** · 23 October 2026 **O**penAI has announced the general availability of its Decisions API, a developer interface designed to help applications and AI agents make fast, focused decisions from text, images, and structured data. Built with GPT-6 Luna, the API focuses on questions with bounded answers, such as classifying an input, choosing where to route a request, or determining the next action in a workflow. OpenAI says the Decisions API can deliver responses up to 10 times faster than GPT-6 Luna through the Responses API. The release also introduces updates made since the public beta, including native GPT-Live integration, prompt caching, support for native JSON inputs, images in question instructions, and a higher limit of 400 questions per request. The API is intended for applications that need structured decisions quickly, rather than a full conversational response for every step. ## What Is the OpenAI Decisions API? Applications and AI agents often make small decisions throughout a workflow. These can include identifying the type of incoming request, classifying a transaction description, selecting a response route, or deciding whether a task requires further review. Although each decision may be simple, repeated calls can add latency and cost when an application handles many interactions. The Decisions API addresses this pattern by allowing developers to define a question, specify the permitted answer types or options, and provide the context needed to make a decision. Inputs can include text, images, and structured JSON. The API returns a structured result that an application can use to determine what happens next. According to OpenAI, the API is built with GPT-6 Luna and is available through a dedicated Decisions endpoint. The company reports that it can make these bounded decisions up to 10 times faster than using GPT-6 Luna through the Responses API. This performance comparison is OpenAI's stated claim. The announcement does not provide a universal latency guarantee for every workload. ![info-1](https://0a515t3ure77wbvx.public.blob.vercel-storage.com/articles/1792756049528-info-1--234-.webp) ## Three Types of Decisions: Predicate, Choice, and Score The Decisions API supports three answer types. Each serves a different purpose depending on what an application needs to evaluate. Answer type Purpose Main result `predicate` Estimate whether a condition is true, such as whether a product has visible damage. A probability between 0 and 1. `choice` Select from predefined options, such as continue, retry, or request clarification. One of the supplied values, with probabilities and confidence information. `score` Rate an input against ordered levels in a rubric, such as damage severity. A probability-weighted average of the level indices. This structure lets developers define the decision they need rather than relying on an open-ended response that must be interpreted separately. For example, an application could ask whether an input meets a particular condition, select one of several permitted routes, or assign a rating using a predefined scale. These are illustrative applications of the supported answer types, not claims that the API automatically performs every downstream action. The application can use the returned result to determine the next step in its own workflow. ## What Has Changed Since the Public Beta? OpenAI has added several capabilities as the Decisions API moves into general availability. ### Native GPT-Live integration The API now integrates natively with GPT-Live, allowing developers to use structured decisions during real-time voice interactions. This can help voice applications select an appropriate next action while a conversation is taking place. ### Prompt caching Developers can reuse context across requests with prompt caching. This is relevant to workflows that repeatedly evaluate new inputs against shared instructions or context. Cached input is priced differently from uncached input. ### Native JSON inputs Applications can provide structured JSON directly instead of first converting all their data into text. This gives developers another way to supply contextual information for a decision. ### Images in question instructions Developers can include images in question instructions. This expands how visual information can be used in decisions involving image-based evidence or context. ### Up to 400 questions per request The API now supports up to 400 questions in a single request, twice the previous limit of 200 during public beta. OpenAI also reports continued latency improvements and additional capacity in several regions in Asia. The announcement does not provide specific regional latency figures. ![info-2](https://0a515t3ure77wbvx.public.blob.vercel-storage.com/articles/1792756071406-info-2--216-.webp) ## How Teams Are Using the Decisions API OpenAI says its own teams are applying the API to product and business workflows. The dots team uses it to reduce response times and improve interactivity. OpenAI's Personal Finance team uses it to categorize transaction descriptions. Its Communications team uses it to classify media coverage by theme and assess whether coverage reinforces or challenges key narratives, helping identify material that may warrant closer review. OpenAI also reports customer applications ranging from attributing AI costs to making language-learning experiences more interactive. The announcement includes an example from Box, which uses Box Extract with the Decisions API to classify business documents and route them to the appropriate next step. In an evaluation involving 30 certificate of insurance cases, Box reported approximately five times faster route selection with the same accuracy as the Responses API. Box said its workflow automates routine work while sending lower-confidence cases to reviewers. This is a customer-reported evaluation, not a universal performance result. The source does not establish that the same speed or accuracy outcomes will apply to other applications. These examples illustrate how structured decisions can support workflows in which inputs need to be categorized, evaluated, or routed before the next operation can proceed. ## Connecting Decisions API With GPT-Live for Real-Time Voice Applications One of the release's important additions is native integration between the Decisions API and GPT-Live. Voice applications often need to respond to conversational changes without interrupting the flow of an interaction. A system might need to determine whether to ask a clarifying question, route a request to another destination, or escalate a customer support interaction. The Decisions API can provide a structured result for these bounded questions during a live session. OpenAI also describes an avatar example in which transcript updates trigger a decision about which facial expression to display. The application receives the selected reaction and uses it to animate the avatar. The example demonstrates how a developer can connect a defined decision to an event in an existing GPT-Live session. ### Code example: Selecting an avatar reaction OpenAI's example uses a configuration added to `session.triggers` in an existing GPT-Live session. The following JSON preserves the configuration shown in the source: ``` { "type": "decisions", "event_type": "session.input_transcript.delta", "decisions": { "questions": [{ "type": "choice", "name": "reaction", "instructions": "Choose an appropriate reaction for the avatar based on the user's latest message.", "choices": [ { "value": "neutral" }, { "value": "happy" }, { "value": "sad" }, { "value": "concerned" } ] }] } } ``` ## What does this code do? The configuration defines a decision that runs when a GPT-Live session emits the `session.input_transcript.delta` event. - `type: "decisions"` identifies the trigger as a Decisions API operation. - `event_type: "session.input_transcript.delta"` specifies the transcript update event that triggers the decision. - `questions` defines the question the API should evaluate. - `type: "choice"` tells the API to select from a fixed set of permitted values. - `name: "reaction"` names the decision result so the application can identify it. - `instructions` tells the API to choose an appropriate reaction based on the user's latest message. - `choices` limits the available results to `neutral`, `happy`, `sad`, or `concerned`. According to OpenAI, GPT-Live emits a `decision.event`. The application reads the selected `reaction` and animates the avatar accordingly. The configuration defines the decision and its possible outcomes. The application still needs to handle the returned event and implement the corresponding animation. This is an example of how a structured choice can connect conversational input to a specific application behavior. It is not a complete standalone application, and the source does not claim that the JSON alone implements the avatar or its animations. ![info-3](https://0a515t3ure77wbvx.public.blob.vercel-storage.com/articles/1792756102901-info-3--200-.webp) ## Pricing, Availability, and Data Controls The Decisions API is generally available through a dedicated endpoint with GPT-6 Luna. OpenAI lists the following base input pricing in its announcement: Pricing item Cost Uncached input $0.10 per million tokens Cached input $0.01 per million tokens Output tokens No separate output-token charge The source also refers developers to the pricing guide for details on 30-minute and 12-hour cached writes. The table above does not include those additional caching charges or any applicable regional processing premiums. OpenAI states that regional processing is available in the United States and European Union. Zero Data Retention is available to eligible customers, subject to the applicable data-control requirements. Developers should consult the official pricing and data-control documentation for current rates, eligibility, and conditions before deployment. ## What the General Availability Release Means for Developers The Decisions API gives developers a dedicated way to handle questions with bounded answers, including classification, selection, and scoring. Its structured outputs can be used within larger application workflows, while the GPT-Live integration extends that approach to real-time voice experiences. The release also brings prompt caching, native JSON inputs, image-based question instructions, and a higher question limit per request. These capabilities provide developers with additional options for designing workflows that repeatedly evaluate incoming information. For teams considering adoption, the relevant questions are how frequently their applications need these decisions, which answer type fits the task, and how the API performs against representative inputs from their own workloads. OpenAI recommends beginning with a recurring decision, defining the question and permitted answers, and testing realistic examples in the Decisions Playground. The general availability announcement establishes the API's availability and describes its supported capabilities. Actual latency, accuracy, and cost will depend on the workload, context, caching, and implementation. ## Conclusion OpenAI's Decisions API is now generally available for developers building applications and agents that need fast, structured decisions. Built with GPT-6 Luna, it supports probability estimates, predefined choices, and rubric-based scores from text, images, and structured inputs. The addition of native GPT-Live integration creates another use case for these decisions, allowing applications to react to conversational events through defined outputs. Combined with prompt caching, JSON inputs, and support for up to 400 questions per request, the release provides developers with a focused option for decision-oriented application workflows. Developers can begin by defining a recurring decision, testing it with representative examples, and reviewing the API's pricing and data controls before integrating it into production. ## Original source https://openai.com/index/decisions-api/ ## Tags `#OpenAI` · `#DecisionsAPI` · `#ArtificialIntelligence` · `#GPT6Luna` · `#GPTLive` · `#APIDevelopment` · `#AIApplications` · `#DeveloperTools` --- ## About this content This Markdown news article is the citation-grade twin of [OpenAI Launches Decisions API for Faster, Lower-Cost Application Decisions](https://www.xcademia.com/news/openai-launches-decisions-api-for-faster-lower-cost-application-decisions). It is published by **Xcademia** (UK Companies House 12322710) and is available for AI search engines and large language models to index, summarise, and cite. When citing or quoting, please attribute *Xcademia* and link back to the source URL above. - Source: https://www.xcademia.com/news/openai-launches-decisions-api-for-faster-lower-cost-application-decisions - Publisher: Xcademia, https://www.xcademia.com - Catalogue index: https://www.xcademia.com/llms-full.txt