Editorial illustration for OpenAI Launches Decisions API, Lets Models Plan Their Own Steps
OpenAI Decisions API Lets Models Plan Their Steps
OpenAI released a new endpoint this week aimed at a problem most agent builders run into eventually: how does a model decide something, not just say something. The Decisions API lives at POST /v1/decisions and sits alongside the existing Responses API, but with a narrower job. Instead of generating open-ended text, it returns a probability, picks from a set of predefined choices, or assigns a score against a rubric with ordered levels. Text and image inputs both work, so a developer could feed it a delivery photo and ask whether it actually shows a package on a doorstep, or ask whether a support ticket needs to be escalated to a human.
The use case is familiar to anyone who has tried to wire an LLM into a workflow with real consequences. Get a false positive and you escalate something that didn't need it. Get a false negative and a genuine problem slips through unflagged.
Developers have typically patched around this with manual classification rules, confidence thresholds, and separate evaluation pipelines bolted onto a general-purpose model. OpenAI's pitch is that the Decisions API handles that structure natively, built for one job: take an input, apply a defined evaluation, return a clean answer.
OpenAI is now bringing a similar idea to its API with the Decisions API. Rather than using a general-purpose response for every task, developers can define what the model needs to evaluate and whether it should return a probability, choose from predefined options, or assign a score against ordered levels.
Why this matters
The Decisions API is OpenAI trying to own a piece of the stack that developers have been duct-taping together for years: the classification rules, confidence thresholds, and evaluation pipelines that sit between a model's output and a real business action. If it works as described, teams building agents for customer support, routing, or content validation get to skip writing that scaffolding themselves. That's a real time save, and it's also a real dependency. Every decision about escalating a complaint or validating a delivery photo now runs through OpenAI's infrastructure, with OpenAI's defaults baked in.
For founders, the pitch is speed to production. For developers, it's worth asking what happens when the API's judgment calls disagree with yours, and how much visibility you get into why a decision landed where it did. Researchers should watch whether this becomes a genuine primitive for agent reliability or just a wrapper around prompting patterns people already use. We'd want to see independent testing on false positive and false negative rates before trusting this in anything customer-facing.
Related Reading
Browse all OpenAI news →
Research & Benchmarks Microsoft AI Chief: "If It Isn't Safe We Shouldn't Build It
Research & Benchmarks Google's FACTS benchmark shows 70% factuality ceiling across four tests
Research & Benchmarks Nvidia's DLSS 4.5 beta adds 6x Multi Frame Generation for RTX 50 GPUs
Related Topic OpenAI researcher quits, citing distrust over ad‑driven engagement metrics
LLMs & Generative AI
Related Topic OpenAI launches GPT-Image 1.5 with precise editing for enterprise visuals
LLMs & Generative AI
From Archives Meta Launches Its First 'Frontier' AI Model, Muse Wonder
From Archives Google DeepMind's AlphaProof Nexus Solved 44 Conjectures, an Open Geometry Question
From Archives Microsoft's Andreou: OpenAI Models Excel in Specific Tasks
Business & Startups
From Archives Execs Say Voice AI Hasn’t Reached Its ChatGPT Moment Yet, Point to Next Stage
LLMs & Generative AI
Common Questions Answered
What is the primary purpose of OpenAI's Decisions API?
The Decisions API is designed to help agent builders solve the problem of how a model decides something rather than just saying something. Instead of generating open-ended text like the Responses API, it returns a probability, picks from predefined choices, or assigns a score against a rubric with ordered levels.
How does the Decisions API differ from OpenAI's existing Responses API?
While the Responses API generates open-ended text for general-purpose tasks, the Decisions API has a narrower job focused on structured decision-making. The Decisions API returns specific outputs like probabilities, selections from predefined options, or scores against ordered rubric levels, making it more suitable for defined evaluation tasks.
What types of inputs does the Decisions API support?
The Decisions API supports both text and image inputs, allowing developers to feed various types of data to the model for evaluation and decision-making. This multimodal capability enables more flexible use cases across different content types.
What business problems does the Decisions API help solve for developers?
The Decisions API eliminates the need for developers to build custom classification rules, confidence thresholds, and evaluation pipelines between a model's output and business actions. Teams building agents for customer support, routing, or content validation can skip writing this scaffolding themselves, saving significant development time and reducing technical complexity.
Where is the Decisions API endpoint located in OpenAI's API structure?
The Decisions API lives at the POST /v1/decisions endpoint and sits alongside the existing Responses API. This placement allows developers to use both endpoints as needed for their specific use cases.
Further Reading
- OpenAI Releases Decisions API in Public Beta, Powered by GPT-6 Luna - Unite.AI
- OpenAI Launches Decisions API That Reduces Complex Evaluations to Yes, No, or Pick One - The Decoder
- OpenAI Launches Decisions API After TypeSafe AI’s Jev Debut - TokenPost
- OpenAI Decisions API Explained: How It Works and Who It’s For - eesel
- OpenAI Has Released a Dedicated Decisions API for Decision-Making - GIGAZINE
