OpenAI Decisions API
Public beta as of October 6, 2026. POST /v1/decisions on gpt-6-luna returns a probability, a choice, or a score. The decisions guide prices that endpoint at $0.10 per million input tokens and no output charge.
OpenAI
Usage-based
$0.10 / 1M input tokens on /v1/decisions. No output charge on that endpoint.
2026-10-07
Use it when the answer is already a type. Do not use it to write the reply.
OpenAI's API changelog for October 6, 2026 says the Decisions API shipped in beta with gpt-6-luna. The guide says it returns typed answers about 10x faster than the Responses API. That speed line is OpenAI's. This directory did not time the two endpoints.
The only model on the guide is gpt-6-luna. The route is POST /v1/decisions. The guide says public beta, with GA expected in the coming weeks. It also names the SDK floors for the examples: Python 3.26.0, JavaScript 7.30.0, Go 3.73.0, Ruby 0.101.0, and Java 4.78.0, or later.
Three fields. model. input, a string or user messages with text and images. questions, each with a type, a unique name, and instructions.
| Type | You get back |
|---|---|
predicate |
probability from 0 to 1 that the condition is true |
choice |
one of your values, plus probabilities and a separate confidence |
score |
a probability-weighted average of level indices, which can fall between levels |
choice is for unordered labels, such as a department. score is for ordered levels, such as severity. The guide's score example uses indices 0, 1, and 2 and shows an illustrative score of 1.1 from probabilities 0.1, 0.7, and 0.2. That JSON is an illustration in the docs, not a measurement from this directory.
Images have to be inline base64 data URLs. Hosted HTTP or HTTPS image URLs and file_id inputs are not supported on this endpoint.
If the next question depends on the first answer, send a second request. Independent questions can share one questions array.
The guide says to include a fallback such as other when your categories do not cover the input, and to set thresholds from labeled examples in your own app. A probability is not a policy until you pick the cutoff.
Use Structured Outputs on the Responses API when you need a JSON object or a written explanation. Use function calling when you need a tool call. The decisions guide draws that line itself.
On gpt-6-luna, the decisions guide says input costs $0.10 per million tokens. You pay only for input tokens. No cache-read, cache-write, or output-token charges on /v1/decisions.
Regional processing premiums and long-context input multipliers still apply. Other gpt-6-luna calls follow the normal model pricing page, not this endpoint's input-only rate.
The guide says Zero Data Retention and HIPAA are available for eligible customers, and that data residency and regional processing are supported in the United States and in Europe (EEA and Switzerland). Eligibility is on OpenAI's data-controls page, not in this note.
The October 6 changelog also says usage tiers were simplified from five to three: Build, Launch, and Grow. The rate-limits page fetched this run lists:
- Free: allowed geography, $100 / month
- Build: $5 in total credit purchases, $500 / month
- Launch: $100 in total credit purchases, $5,000 / month
- Grow: $500 in total credit purchases, $200,000 / month
Those are monthly usage limits, not the per-token price. Check the limits page in the console for the RPM and TPM on your tier. This note does not copy a per-model cap that was not on the page.
- Guide: https://platform.openai.com/docs/guides/decisions
- Changelog, Oct 6: https://platform.openai.com/docs/changelog
- Rate limits: https://platform.openai.com/docs/guides/rate-limits