OpenAI Decisions API
OpenAI Decisions API: what it is, what it costs and its limits
Short answer
The OpenAI Decisions API is an endpoint (POST /v1/decisions) that answers typed questions about text or images instead of generating text. It supports predicate (probability), choice and score questions, runs only on gpt-6-luna, costs $0.10 per 1M input tokens with no output charge, and has been in public beta since October 2026.
OpenAI launched the Decisions API to give developers fast, typed answers for classification, routing and prioritization. OpenAI says it returns answers about 10x faster than the Responses API. This page sums up what the official guide says, what it doesn't cover, and how it compares to a full decision layer such as ClassifierHub.
Everything below is taken from OpenAI's public documentation as of the date at the bottom of the page. The API is in beta, so check the official guide before you budget or build.
Get Early Access to ClassifierHub: 2× credits in your first paid month.
How a Decisions request works
A request has three fields: model (currently only gpt-6-luna), input (a text string or user messages with text and images) and questions (an array of typed questions, each with a unique name). The response contains an answers array that echoes each question's name.
- predicate: checks a condition and returns probability, an estimate from 0 to 1 that it is true.
- choice: picks one value from the choices you supply and returns the choice, a probabilities array and a confidence field.
- score: rates the input against ordered levels and returns a probability-weighted average of the level indices, so 1.1 can fall between levels 1 and 2.
- Multiple independent questions can share one input in a single request. Questions that depend on an earlier answer need separate requests.
curl https://api.openai.com/v1/decisions \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-6-luna",
"input": "I was charged twice for my order.",
"questions": [{
"type": "choice",
"name": "department",
"instructions": "Which department should handle this complaint?",
"choices": [
{"value": "billing", "description": "Payments, invoices, and refunds."},
{"value": "technical", "description": "Problems using the product."},
{"value": "other", "description": "Requests outside these categories."}
]
}]
}'
# -> answers[0]: { "name": "department", "choice": "billing",
# "confidence": ..., "probabilities": [...] }OpenAI Decisions API pricing
With gpt-6-luna, the Decisions endpoint costs $0.10 per 1M input tokens. There are no output-token, cache-read or cache-write charges. Regional processing premiums and long-context input multipliers still apply, and these rates cover only /v1/decisions: other gpt-6-luna calls use standard model pricing.
Because you pay per input token, the cost of a decision grows with the size of what you send. A short message costs a tiny fraction of a cent; a long email thread, document or large JSON record costs proportionally more, and images add their own input tokens.
Limits to know before you build
- Beta status: OpenAI describes the API as public beta, with general availability expected in the coming weeks. Behavior and pricing can change at GA.
- One model: gpt-6-luna is the only supported model, so there is no fallback provider inside the endpoint.
- Image inputs must be inline base64 data URLs. Hosted image URLs and file_id inputs aren't supported.
- Typed answers only: for extracted fields or written explanations, OpenAI points you to Structured Outputs or function calling in the Responses API.
- It is an API primitive. Templates, saved decisions, batch jobs over spreadsheets, per-team credit budgets and an MCP server for AI agents are things you build yourself.
- Compliance: ZDR and HIPAA are available for eligible customers, and data residency is supported in the United States and Europe (EEA and Switzerland).
When the Decisions API is a good fit
It's a strong choice if you already run on OpenAI, your inputs are short text or you need image understanding, and you have engineers to build the surrounding workflow. For short-text decisions at very high volume, its raw per-call price is hard to beat.
If you want decisions ready to use from AI agents, spreadsheets and no-code tools, with a fixed monthly price per decision instead of token math, a decision layer like ClassifierHub gets you there without building the plumbing. See the side-by-side comparison for details.
Flat price per decision, MCP for every AI tool, Excel built in.
Compare with ClassifierHubFrequently asked questions
Related guides
Last updated . ClassifierHub is an independent product built on top of the Jev decision model, accessed through OpenRouter. It is not affiliated with or endorsed by TypeSafe or OpenRouter.