Decisions API Is Now GA: Up to 10x Faster for Small, Bounded Decisions!

Hi everyone, it's Shii-chan! Today OpenAI has news about an API that returns small decisions fast and cheaply!

OpenAI News openai.com

What was announced?

In OpenAI's News, the Decisions API was announced as generally available. It is built on GPT-6 Luna and focuses its intelligence on questions with bounded answers, so applications can make near real-time decisions from text, images, and structured inputs.

Why it matters

Applications and agents make lots of small decisions in a workflow, such as classifying an input, routing a request, or picking the next action. Each one is simple, but at scale their latency and cost add up.

The Decisions API targets exactly that. According to OpenAI, it is up to 10x faster than GPT-6 Luna through the Responses API.

What changes

Here is what was added since the public beta:

• Native GPT-Live integration
• Prompt caching for reused context
• Inference residency in the US and EU
• Native JSON inputs and images in question instructions
• Up to 400 questions per request, twice the previous limit
• More capacity in several Asian regions, with improved latency

Inside OpenAI, teams use it for things like categorizing transaction descriptions and classifying press coverage by theme and narrative impact.

Dive Deep

You define a question and its allowed answers, then provide context as text, images, or native JSON. There are three answer types:

• predicate: estimates the probability (0 to 1) that a condition is true, such as whether a product has visible damage
• choice: selects one of your predefined options, such as whether an agent should continue, retry, or ask for clarification. The response also includes probabilities and confidence
• score: rates an input against ordered levels in a rubric. The result is the probability-weighted average of the level indices

With GPT-Live, you add a configuration to session.triggers in an existing session. The session emits a decision.event, and your app reads the selected reaction, for example to change an avatar's expression. Another example is a support agent deciding mid-conversation whether to ask a clarifying question, route a request, or escalate.

Pricing:

• Input: $0.10 per million tokens
• Cached input: $0.01 per million tokens
• See the pricing guide for 30-minute and 12-hour cached writes

To start, pick one decision your app makes often and define the question, the answers, and the context. You can try it with your own text or images in the Decisions Playground.

Wrap-up

• The Decisions API is GA, built on GPT-6 Luna
• It focuses on bounded questions and is up to 10x faster than GPT-6 Luna through the Responses API
• Three answer types: predicate, choice, and score
• Adds GPT-Live integration, prompt caching, US and EU inference residency, and 400 questions per request
• Input costs $0.10 per million tokens

This one is for developers running lots of classification and routing, or building voice agents!