OpenAI Decisions API: faster typed answers for apps

OpenAI is adding a more opinionated tool to its stack: the Decisions API. Instead of asking a model to write something, you ask it to pick a category, estimate a probability, or score an input against ordered levels. OpenAI says it can do that about 10x faster than the Responses API, and it’s meant for apps that need a decision more than a paragraph.

The API is in public beta and OpenAI expects to move it to general availability in the coming weeks. Right now, the only model available is gpt-6-luna, and requests go through a dedicated POST /v1/decisions endpoint. The supported input can be plain text, user messages, or a mix of text and images.

The framing is pretty practical. A predicate question checks whether something is true, like visible damage on a product photo. A choice question picks one value from a fixed set, such as billing, technical, shipping, or other. A score question ranks something against ordered levels, which is handy for things like issue severity. Each question gets a name, and the API sends that name back in the answers array so applications can route the result cleanly.

OpenAI is also drawing a line around when not to use it. If the job is to generate an object that follows your own JSON schema, it points developers to Structured Outputs in the Responses API. If the model needs to ask for a tool call with arguments, that’s still function calling. Decisions is for classification, routing, filtering, and review thresholds.

There are some sharp edges. Images must be sent as inline base64 data URLs; hosted image URLs and file_id inputs are not supported here. OpenAI also says the SDK examples require newer versions across Python, JavaScript, Go, Ruby, and Java. Pricing is different too: with gpt-6-luna, input costs $0.10 per 1M tokens, with no cache-read, cache-write, or output-token charges. The endpoint supports Zero Data Retention and HIPAA for eligible customers, and data residency plus regional processing in the United States and Europe.