OpenAI launches the Decisions API for fast AI app decisions
OpenAI's Decisions API is now generally available, letting apps and AI agents make small decisions 10x faster at low cost.
OpenAI made its Decisions API generally available to all developers starting October 22, 2026. The API is built with the GPT-6 Luna model and designed for one job: helping apps and AI agents make small decisions fast. Think prioritizing customer complaints, or deciding whether a transaction needs to be flagged.
1. How it differs from a normal API
Normally, an AI app generates a full answer for every question. That is slow and expensive when the question is just "pick one of these options." The Decisions API takes a different approach: the developer defines a question and a set of allowed answers, and the API returns one structured answer.
Three answer types are supported: a probability (for example, 0.82), a choice from a fixed list, or a score against a defined rubric. The app uses that answer to decide what happens next, without processing generated text.
2. How fast and how cheap
OpenAI claims the Decisions API is up to 10x faster than GPT-6 Luna through the Responses API, the usual way to call the model. That figure comes from OpenAI itself, so it is a vendor claim, not an independent measurement.
Pricing is straightforward: $0.10 per million input tokens, and $0.01 per million for cached input. There are no output token charges at all. A million tokens is roughly the size of several textbooks, so the cost of a single decision is tiny.
3. What is new since the beta
The Decisions API previously ran as a public beta. Since then, OpenAI added several capabilities:
- Native GPT-Live integration, so voice apps can react as people speak
- Prompt caching to cut costs when context is reused
- Native JSON inputs for structured app data
- Images in question instructions, to compare multiple images against one reference
- Up to 400 questions per request, double the previous limit
Regional processing is now available in the US and EU, plus added capacity in several Asian regions.
4. Azure joined on day one
Microsoft brought the Decisions API to Microsoft Foundry on October 22 as well. Azure developers can call a dedicated endpoint using OpenAI SDKs or REST, with GPT-6 Luna as the model. Companies already inside the Azure ecosystem do not need to leave their existing setup to use it.
5. Who should care
The Economic Times reported that the launch puts OpenAI in competition with decision models such as Jev from TypeSafe, one of the earliest models built specifically for these tasks. The AI decision market is still young, but the direction is clear: not every model call needs to generate a long answer. Most of an AI agent's work is actually small choices repeated thousands of times.
In Malaysia, this matters for teams building customer chat, document review, or back-office automation. Decisions priced this low can sit directly inside a workflow without straining the budget.
What it means for you
If you are not a developer, this news may look technical. But it shapes how the AI apps you use will respond faster and cost less in the coming months.
- Developers: try the Decisions Playground to test decision questions before writing code. One frequently repeated decision in your app is the best candidate.
- Business owners: ask your app vendor whether they use APIs like this for classification and routing. Operating costs can drop significantly.
- Everyday users: expect more responsive voice and chat apps, because small decisions no longer wait on a full model response.
- Learn on your own: tools like AI Tools can help you understand today's AI tool categories before picking one for your work.
Sources
- Build more responsive apps with the Decisions API (OpenAI)
- Decisions API is now generally available with new updates (OpenAI Developer Community)
- Azure OpenAI Decisions API: Now generally available in Microsoft Foundry (Microsoft)
- OpenAI enters Jev's turf with a decision API for AI agents (The Economic Times)