OpenAI opened its Decisions API to all developers in public beta on 6 October 2026, in a post on the OpenAI Developer Community. The API runs on GPT-6 Luna and returns typed answers rather than generated text: a probability, a choice from a fixed list, or a score.
OpenAI also published a price. Its docs say that with gpt-6-luna, input costs $0.10 per 1M tokens, and that you pay only for input tokens: there are no cache-read, cache-write, or output-token charges. (Update, 28 October 2026: OpenAI’s current Decisions guide says prompt cache hits receive discounted input pricing and cache writes are charged, at the standard gpt-6-luna prompt caching rates; output tokens are still not charged.) When we covered the limited preview on 1 October, no price had been published.
Business Pill · SENDING EACH QUESTION TO THE RIGHT MODEL
A one-minute explainer of model routing: an application sends each request to the model that fits it, rather than sending everything to the largest one. It teaches the general idea only and says nothing about any company in this story.
The key insight: The price is Luna’s input price, unchanged; what disappears is the output bill. OpenAI’s pricing page lists gpt-6-luna at $0.10 per 1M input tokens and $0.50 for output, and the Decisions endpoint charges the $0.10 input rate with no output charges (cache writes are now charged at standard rates; see the 28 October update), with regional and long-context multipliers on top. As we read it, the product is priced on what it reads, because what it returns is a number or a label.
The docs describe a dedicated POST /v1/decisions endpoint. A request has three parts: the model, the input, and the questions. They say gpt-6-luna is the only model currently available, and that OpenAI expects the API to reach general availability in the coming weeks.
The announcement says the Decisions API makes decisions up to 10x faster than GPT-6 Luna through the Responses API. The docs put it as typed answers about 10x faster than the Responses API. We have not tested either claim.
Inputs can be text, images or both. The docs say images must be inline base64 data URLs, and that hosted image URLs and file_id inputs are not supported by this endpoint.
gpt-6-luna prices per 1M tokens in US cents: standard short-context rates from OpenAI’s pricing page (10 input, 1 cached input, 12.5 cache writes, 50 output) against the Decisions API rate in OpenAI’s docs, which charges 10 for input and nothing for cache reads, cache writes or output. Both read on 7 October 2026. The docs say regional and long-context multipliers apply to the Decisions rate.
What Was Posted on X
ICYMI ๐: OpenAI Platform now has a “Decisions” API playground.
Users can test OpenAI’s decision model with the “Predicate,” “Choice,” and “Score” options, which were released recently.
Probability True ๐ค
— ๐จ AI News | TestingCatalog (@testingcatalog) Oct 20 00:07
TestingCatalog posted this at 00:07 UTC on 20 October, saying the OpenAI Platform now has a Decisions API playground where users can test the predicate, choice and score options. OpenAI’s Decisions guide points developers to the Playground to experiment with questions and inputs before writing code, and says the API is in public beta.
Three Answer Types
A predicate returns an estimate from 0 to 1 that a condition is true. A choice returns one of the developer’s supplied values. A score rates an input against ordered levels.
The docs say a score is the probability-weighted average of the level indices, so it can fall between levels. Their example uses three severity levels numbered 0 to 2, where probabilities of 0.1, 0.7 and 0.2 produce a score of 1.1.
OpenAI tells developers to use labeled examples from their own application to set thresholds for routing, filtering or review, based on the cost of false positives and false negatives.
What the Price Changes
OpenAI’s API pricing page lists gpt-6-luna at standard short-context rates of $0.10 per 1M input tokens, $0.01 cached input, $0.125 for cache writes and $0.50 for output. The Decisions rate matches that input price and drops the output charge; under OpenAI’s current guide, cache hits are discounted and cache writes are charged at the standard rates.
The docs add that regional processing premiums and long-context input pricing multipliers apply, and that other requests using gpt-6-luna follow the applicable model and processing-tier pricing. They also say the API supports Zero Data Retention and HIPAA use for eligible customers, with data residency in the United States and Europe.
The First Developer Reactions
In the announcement thread, one developer posting as jefff said they ran quick tests against Jev on a few hundred questions from word games and one conversational game. They reported that Jev was more accurate on nuanced judgment calls and faster, and that Luna cost about 2–3× as much for the same questions.
The user added their own caveat: “treat it as a strong early signal rather than a general benchmark.” Another user replied that image understanding is something Jev does not support yet. These are individual posts, not OpenAI statements, and we have not run any comparison.
Update, 20 October: A Playground and a Voice Hook for Decisions
OpenAI’s API changelog entry dated 15 October 2026 says OpenAI “Added native Decisions integration in GPT-Live,” so that an app “can act while your user is speaking or react to what Live says.” Developers can “Let Live call Decisions directly through native delegation, or trigger decisions as input or output transcript deltas arrive,” the entry says.
The Decisions guide tells developers to “Try the Decisions API in the Playground to experiment with questions and inputs before writing code.” TestingCatalog posted on 20 October 2026 that the OpenAI Platform now has a Decisions API playground where users can test the predicate, choice and score options. The guide still says the Decisions API “is in public beta” and that gpt-6-luna “is the only model currently available.”
The changelog entry and the guide text read for this Update give no new price and no usage figures.
As we read it, both changes make the endpoint easier to try without new code: a playground for testing questions, and a hook for voice apps to call it mid-conversation. They land in the same week OpenAI added shared, co-edited Slides to ChatGPT, which we covered in an Update to our 17 October piece on ChatGPT controlling Chrome tabs.
The Structural Read
The answer format is what makes input-only billing workable. A Decisions call returns a probability, one of the developer’s own values, or a score, not prose, so there is little output to meter. That is our reading of the docs’ three answer types.
OpenAI puts the threshold on the customer. The docs tell developers to set routing, filtering or review thresholds from their own labeled examples, weighing the cost of false positives and false negatives. The API supplies the probability; the cut-off is the developer’s decision.
The beta has limits written into the docs: one model, gpt-6-luna; images only as inline base64; separate requests for questions that depend on an earlier answer; and regional and long-context multipliers on top of the base rate.
OpenAI API docs, Decisions guide
“Use labeled examples from your application to set thresholds for routing, filtering, or review.”
Three Implications
INPUT-ONLY BILLING The docs price Decisions at $0.10 per 1M input tokens on gpt-6-luna with no output-token charges (the current guide charges cache writes at standard rates); the pricing page lists Luna’s standard output at $0.50.
THRESHOLDS ARE THE DEVELOPER’S JOB OpenAI tells developers to calibrate thresholds on their own labeled examples, based on the cost of false positives and false negatives.
GA EXPECTED IN THE COMING WEEKS The docs say OpenAI expects general availability in the coming weeks, with gpt-6-luna as the only model currently available.
The framework’s starting point: “It is simultaneously the unit of cognition the model produces, the unit of compute the data center serves, the unit of price the lab charges, and the unit of value the enterprise extracts.”
As we read it, the Decisions API pulls that bundle apart: OpenAI’s docs charge for input tokens only, with no output-token charges, for an endpoint that returns a probability, a choice or a score, so the price follows the tokens the model reads.
What Is Not Established
We read the announcement thread, the Decisions guide and the pricing page. OpenAI has published no accuracy figure for the API in the documents we read, and the 10x speed figure is its own.
The developer comparison in the thread is one person’s informal test. We did not contact OpenAI.
Business Engineer Framework
The Map of AI — Where Model Labs Sit
The Map of AI places more than 200 companies across nine layers of the stack, from silicon to application. It shows where model labs sit relative to the developers and software companies that build on their APIs.
OpenAI’s Decisions API is now in public beta for all developers, on gpt-6-luna only, with general availability expected in the coming weeks. It is billed at Luna’s $0.10 per 1M input rate with no output or cache charges, and it returns probabilities, choices or scores instead of text.
A note on sourcing. We read OpenAI’s announcement thread on the OpenAI Developer Community in full, including the replies, and the Decisions guide and pricing page in OpenAI’s API docs. The speed figure is OpenAI’s and the developer comparison is one user’s informal test; we have run no tests and did not contact OpenAI. The 20 October Update rests on OpenAI’s API changelog entry of 15 October 2026, the current Decisions guide, and TestingCatalog’s post of 20 October 2026. Nothing here is a forecast, and nothing here is financial or investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
