On 6 October 2026 the OpenAI API changelog says the Decisions API is released in beta with gpt-6-luna. The Decisions guide calls it a public beta and says OpenAI expects general availability in the coming weeks. The only model listed is gpt-6-luna, on POST /v1/decisions.
The DevDay 2026 recap, dated 29 September 2026, called the API a limited preview, with a broad release planned in the coming days. It described text or image context and finite pre-defined answers for classifying content, routing a request, or choosing an agent’s next action.
What a request returns
A request has model, shared input (a string, or user messages with text and images), and questions. Each question has a unique name, echoed in answers.
| Type | What the guide says you get |
|---|---|
predicate |
A probability from 0 to 1 that a condition is true |
choice |
One of the values you supplied |
score |
The probability-weighted average of ordered level indices |
choice and score also return a probability per option and a confidence field. The guide says to set thresholds from your own labeled examples.
Level indices start at 0, so a score can fall between levels. The guide’s illustrative example uses probabilities 0.1, 0.7, and 0.2 and shows 1.1.
The API reference says a choice may be a string or a boolean, and that a string and a boolean with the same text stay distinct. It also documents refusal: the host may decline one question without disclosing its refusal score.
Independent questions can share one questions array. The guide says a follow-up that depends on an earlier answer is a separate request.
Images are inline base64 data URLs. The guide says hosted HTTP or HTTPS URLs and file_id inputs are unsupported. The reference says external URLs and file IDs are unsupported, and that detail may be low, high, auto, or original, defaulting to auto.
The price on the decisions guide
With gpt-6-luna, the guide sets input at $0.10 per 1M tokens and says there is no cache-read, cache-write, or output-token charge. Regional processing premiums and long-context input multipliers apply. The decisions page does not print those multipliers. This rate is for /v1/decisions. Other gpt-6-luna requests follow that model’s processing-tier prices.
The 22 September changelog lists standard GPT-6 Luna prices, for prompts up to 272K input tokens, as $0.10 input, $0.01 cached input, and $0.50 output per 1M tokens. The decisions input figure matches that input price.
The reference still returns usage with input, output, and cache token counts.
The changelog says typed answers come back 10x faster than the Responses API. The guide uses 10x, and one sentence says “about 10x faster.”
Retention, residency, and the same week’s platform notes
The data controls row for /v1/decisions reads: training, No; abuse monitoring, 30 days; application state, “None, see below for exceptions”; Zero Data Retention, “Yes, see below for limitations”; Private Retention with PSP and Safety Retention, “Pending confirmation.”
The page says default abuse-monitoring logs last up to 30 days, and that eligible customers can use Zero Data Retention with the limitations listed there. Prompt caching may keep encrypted key/value tensors on the local GPU until a 24-hour expiration. Image inputs are scanned for CSAM on submission, and a potential CSAM detection is retained for manual review even when Zero Data Retention, Modified Abuse Monitoring, or Private Retention with PSP is enabled.
The page says the API is eligible for HIPAA under an executed OpenAI Business Associate and Healthcare Addendum, subject to account configuration. 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 Europe (EEA and Switzerland). The regional table lists /v1/decisions storage in every region on that table, and processing in the United States and Europe (EEA and Switzerland).
On 5 October the changelog added an in-product flow under Organization settings, General. Admins of eligible organizations can accept the standard Business Associate Agreement and enable HIPAA compliance support. The entry points to the Help Center for eligibility, covered services, and configuration.
The 6 October changelog also simplifies API usage tiers from five to three: Build, Launch, and Grow, with automatic upgrades as total credit purchases reach each minimum. The rate limits page calls those the three paid tiers: Build at $5 purchased and a $500 monthly usage limit, Launch at $100 and $5,000 a month, Grow at $500 and $200,000 a month. Free remains, at $100 a month, for a user in an allowed geography.
For a probability, a fixed-set choice, or a rubric score, call POST /v1/decisions on gpt-6-luna and take $0.10 per 1M input tokens from the decisions guide. Read the data-controls row before regulated content.
The Campfire
No commentsNobody has pulled up a log by this one yet. Be the first to say what you make of it.
Held for the desk. It appears after a look.