AI Coding Tools

OpenAI launches Decisions API to speed routine choices in apps and agents

OpenAIOct 26, 2026

OpenAI released its Decisions API for general use, offering structured choices from text, images, and JSON with input-token-based pricing.

OpenAI made its Decisions API generally available on October 22, 2026, as a dedicated endpoint powered by GPT-6 Luna. The company describes the service as a way for applications and agents to handle bounded decisions—such as classifying an input, routing a request, or selecting a next action—without using a broader response for every step.

Developers specify a question, permitted answers, and supporting context. The API accepts text, images, and native JSON, and returns one of three result types: a probability for a yes-or-no condition, a choice from supplied options, or a score against ordered levels. OpenAI says it can handle as many as 400 questions in one request and can run up to 10 times faster than GPT-6 Luna through the Responses API. Those speed comparisons are the company’s stated performance claims.

The launch includes a native integration with GPT-Live for decisions during voice interactions, along with prompt caching and inference residency in the United States and European Union. OpenAI lists a price of $0.10 per million input tokens and $0.01 per million cached input tokens; output tokens are not charged. Zero Data Retention is available to eligible customers.

OpenAI cites early uses inside the company and at customers including Box, Datadog, Ramp, Sierra, and Speak. The reported results vary by task: Box said its evaluation of 30 insurance-certificate cases produced roughly five-times-faster route selection with the same accuracy as the Responses API. Sierra reported latency below 280 milliseconds at the 95th percentile, while Speak said its GPT-Live integration reduced time to the next action from 1.4 seconds to 0.13 seconds in testing. These are customer-reported evaluations, not guarantees for other workloads. For developers, the API’s fixed answer formats are intended to make frequently repeated routing, scoring, and classification steps easier to integrate into application logic.

Industry Impact

Faster, structured application decisions

Enterprise software, insurance workflows, voice applications

Overview

The generally available API gives developers a dedicated way to handle repeated classification, routing, and scoring tasks with fixed answer formats. Its claimed speed advantage and input-token pricing may reduce latency and simplify integration compared with using a broader response for each decision. For product managers, it offers a focused building block for application workflows; developers can test it against existing implementations. Customer results are task-specific and self-reported, so teams should validate accuracy, latency, and cost on their own workloads before adopting it broadly.

Key Impact Points

Fixed outputs can make routing, classification, and scoring easier to connect to application logic.

OpenAI says the API can run up to 10 times faster than GPT-6 Luna through the Responses API; this is a company performance claim.

Input-token pricing, cached-input discounts, and no output-token charge give teams a distinct cost model to evaluate.

Opportunities

Operationalize bounded AI decisions

Enterprise applications, insurance processing, voice workflows

Organizations with repeated routing, classification, or scoring steps can evaluate the Decisions API as a narrowly scoped component in existing applications and agent workflows. Its text, image, and JSON inputs and supplied-answer formats may suit cases where downstream systems need a predictable choice rather than free-form text. The GPT-Live integration also creates an option for decisions during voice interactions. Teams should compare performance with their current approach using representative cases, including accuracy and tail latency, and account for caching, residency, and data-retention requirements. The published pricing supports a usage-cost assessment, but the article gives no evidence of customer revenue or guaranteed savings.

Automate bounded workflow decisions
Problem: Applications often need repeated classification, routing, or scoring, while a broader response may be unnecessary for each step.
Solution: Evaluate the Decisions API for fixed-format choices using text, images, or JSON as input.
Target Users: Application developers and product teams building enterprise workflows or voice interactions.
Business Value: Potentially simpler integration and lower decision latency; customer-reported outcomes are specific to their tests and are not guarantees.
Revenue Potential: OpenAI charges $0.10 per million input tokens and $0.01 per million cached input tokens, with no output-token charge. The article does not quantify customer revenue or savings.

Reference link

https://openai.com/index/decisions-api

Image Source Link

https://techcrunch.com/wp-content/uploads/2026/10/Screenshot-2026-10-22-at-12.17.11-PM-1.png

Community Discussion (0)

No discussions yet. Be the first to start the conversation!