OpenAI Decisions API: 10x Faster AI Decisions
OpenAI's Decisions API is now available to all developers, letting apps use AI to make quick decisions up to 10 times faster than before. Here's what the new tool does, how it works, and why it matters for the apps you use every day.
Image: OpenAI via Wikimedia Commons (CC0)
What Is the OpenAI Decisions API?
Most of the time when you talk to an AI chatbot, it writes back a full paragraph. That's great for conversation, but a lot of apps don't need a paragraph. They need a quick answer to a simple question: Is this message spam? Which support team should handle this ticket? What should the AI do next?
The OpenAI Decisions API is built for exactly those moments. Instead of generating a long reply, it reads your input and returns a short, structured answer: a probability, a choice from a list you defined, or a score. It is a dedicated endpoint (/v1/decisions) powered by OpenAI's GPT-6 Luna model, focused on one job: making small decisions, extremely fast.
OpenAI first showed the Decisions API in a limited preview at its DevDay developer event on September 29, 2026, then ran a public beta starting October 6. On October 22, 2026, the company opened it to all customers.
How the OpenAI Decisions API Works
Here's the key idea: a normal AI model spends time and money writing out words. If all you need is a yes/no answer or a pick from five options, generating a paragraph is wasteful. The Decisions API skips the paragraph entirely.
You give it three things:
- A question — for example, "Is this customer message urgent?"
- The allowed answers — a fixed set you define ahead of time, like urgent, normal, or spam
- The context — text, an image, or structured data in JSON format
The API reads the context and returns one of three answer types:
The Three Answer Types
- Predicate: a probability between 0 and 1 that a condition is true. For example, there's a 0.92 probability the product in this photo has visible damage.
- Choice: one of your predefined options, plus confidence ratings. For example, an AI assistant should ask for clarification here, not continue or retry.
- Score: a rating against your own rubric — for instance, the severity of a problem on a scale you define.
A single request can carry up to 400 questions — twice the limit from the beta period — so one call can triage hundreds of items at once.
Why 100 Milliseconds Matters
Speed is the headline. According to OpenAI, the Decisions API makes decisions up to 10 times faster than asking GPT-6 Luna the same questions through the standard Responses API.
Nikunj Handa, OpenAI's API product lead, told TechCrunch the API shrinks response times "from one to three seconds to 100, 120, 180 milliseconds." In a tenth of a second, an app can decide what to do next without making you wait.
Real customers are already seeing the difference:
- Speak, the language-learning app, cut the time between actions in its AI tutor from 1.4 seconds to 0.13 seconds — making conversations feel natural instead of laggy.
- Box, Datadog, and Ramp tested the API during the preview, using it to replace slower model calls with higher-throughput decisions.
- Inside OpenAI, the team behind Dots (the AI voice assistant) used it to make voice conversations respond faster, the finance team uses it to categorize transactions, and the communications team uses it to classify news coverage.
Handa says the next target is around 60 milliseconds per decision — fast enough for "an API call for every single frame in a video."
What It Costs
The Decisions API is priced at $0.10 per million input tokens (the chunks of text and images the model reads). There is no charge for output tokens — because the API barely generates any — and cached input costs only $0.01 per million tokens. You pay for what the model reads, nothing more.
The API also supports prompt caching for reused context, native JSON inputs, image inputs in question instructions, and a native integration with GPT-Live for real-time voice apps. OpenAI says regional processing is available in the US and EU, Zero Data Retention is available to eligible customers, and the API is HIPAA eligible for healthcare use.
The Race Against TypeSafe's Jev
OpenAI didn't invent this idea. The category was created by TypeSafe, a startup whose Jev model acts as a super-fast "System One" classifier — and quickly became a $7.6 billion company.
Handa was refreshingly honest about it. He told TechCrunch that when his team saw Jev, colleagues said "wow, we have like a million use cases for this," and the team sprinted to build their own version. Rather than training a brand-new decision model, OpenAI simply reconfigured GPT-6 Luna to speed up its inference and lower the cost: "right now, it's just plain old Luna that gets the job done."
Independent benchmarks suggest OpenAI is still catching up on pure decision quality — the AIM System One Index and the Artificial Analysis Decision Index currently rank newer dedicated models like Liquid's D1 Ultra and Base Lab's Elephant-1 ahead of OpenAI's offering. OpenAI's advantage is multimodality: "Luna is extremely powerful and has been showing better calibrations" for image classification, Handa said.
What This Means for You
You don't need to be a developer to feel this. The Decisions API powers the invisible layer of apps:
- Faster apps: support chats, spam filters, and content moderation can react almost instantly instead of pausing for a second or two.
- Cheaper AI: because the API skips writing long replies, developers pay far less — which can mean cheaper features for you.
- Smarter assistants: voice assistants and AI agents can make dozens of micro-decisions per second — choosing when to speak, when to listen, and what to do next — without lag.
- More private options: smaller, focused models also run locally on devices (Ollama now supports decision models), so some of this could eventually happen without sending your data to the cloud at all.
The big picture: AI is moving beyond writing paragraphs. The next wave is AI that decides — quickly, cheaply, and quietly inside the apps you already use. OpenAI's Decisions API is a clear sign that the race is on.
For more AI news and coverage of the tools developers are building with, keep reading Top Tech Today — and check the latest technology news for what's next.
Sources: TechCrunch, OpenAI
Discussion
// Sign in with Google to join the discussion (Settings → Comments to change who can comment)