OpenAI made the Decisions API generally available on October 22, 2026. In its announcement, OpenAI says the API is built with GPT-6 Luna and focuses the model on questions with bounded answers, from text, images and structured inputs, up to 10x faster than GPT-6 Luna through the Responses API. It costs $0.10 per million input tokens, and OpenAI says "there are no output-token charges."

#What changed?

The Decisions API answers narrow questions instead of writing free text. OpenAI lists three answer types:

  • Predicate: "Estimate the probability that a condition is true, such as whether a product has visible damage." The answer is a probability from 0 to 1.
  • Choice: "Select one of your predefined options, such as whether an agent should continue, retry, or ask for clarification."
  • Score: "Rate an input against ordered levels in a rubric, such as the severity of damage." The docs say a score question can use 2 to 10 ordered levels.

Requests go to a dedicated endpoint. The docs say: "Send requests to POST /v1/decisions using gpt-6-luna, the only currently supported model."

OpenAI's docs list these additions since the public beta:

  • More questions: "Ask up to 400 questions per request, up from 200 during the public beta."
  • JSON input: "Pass structured data directly with an input_json content part."
  • Images: hosted image URLs and inline base64 images, plus images in question instructions.
  • Conversation history: "Evaluate messages alongside function calls, custom tool calls, and their outputs."
  • Prompt caching for reused context.
  • Regional processing in the US and EU.
  • Organization usage reporting and Private Link for private connections.
  • Native GPT-Live integration. OpenAI's changelog dates this to October 15, 2026, a week before general availability.

OpenAI also says it has improved latency, "including additional capacity in several regions in Asia." The docs note: "Files, audio, and item references aren’t supported."

With the GPT-Live integration, OpenAI says, "Live calls Decisions directly," so apps can "make low-latency decisions while your user is still speaking."

#Who can use it and when?

It is generally available from October 22, 2026, and TechCrunch says it is open to all customers. OpenAI's changelog shows the API entered beta on October 6, 2026, and TechCrunch reports it was announced at OpenAI's developer day on September 29.

The docs say: "The Decisions API supports Zero Data Retention (ZDR) and HIPAA use for eligible customers." Regional processing "is supported in the United States and Europe (EEA + Switzerland)."

Pricing, as published (per 1 million tokens):

Item

Price

Uncached input

$0.10

Cached input

$0.01

Output

Not charged

30-minute cache write

$0.125 (standard gpt-6-luna rate)

12-hour cache write

$0.25 (standard gpt-6-luna rate)

The docs say "cache writes are charged" and that "Cache-read and cache-write rates match the standard gpt-6-luna prompt caching rates." The write prices above come from the gpt-6-luna row of OpenAI's pricing page, which does not name the Decisions API. The docs add: "Regional processing premiums and long-context input pricing multipliers apply." To get cached pricing, the reusable prefix "must contain at least 1,024 tokens."

#How to get it

  1. Send requests to POST /v1/decisions with the model gpt-6-luna, per OpenAI's Decisions guide.
  2. Pick an answer type for each question: predicate, choice or score.
  3. Put shared context in a reusable prefix of at least 1,024 tokens to qualify for cached input pricing.
  4. For voice apps, use the native GPT-Live integration. The voice guide says to "Configure session.triggers for input or output transcript deltas."

#Who is using it?

On its announcement page, OpenAI says: "The dots team is using it to reduce response times and improve interactivity." For background on Dots, see our OpenAI Dots launch coverage. OpenAI's Personal Finance team uses it to categorize transaction descriptions, the page says.

OpenAI's page also quotes customers. These are their claims:

  • Box says it saw "~5x faster route selection on 30 certificate of insurance cases, with the same accuracy as the Responses API."
  • Datadog says that against available frontier models, "it delivered 40x faster p95 with comparable accuracy."
  • Ramp says the API made AI cost attribution 4x faster and 30% cheaper than GPT-6 Luna, without losing accuracy.
  • Sierra says it runs "at under 280 ms p95."
  • Speak says the API's GPT-Live integration cut time to next action by over 90%, from 1.4 seconds to 0.13 seconds.

#How does it compare?

TechCrunch frames the launch as competition with TypeSafe, whose Jev model it says "created a category for deep learning models that act as super classifiers." It reports that public benchmarks, the AIM System One Index and the Artificial Analysis Decision Index, tend to favor Liquid's D1 Ultra and Base Lab's Elephant-1. By TechCrunch's account, Jev ranks seventh and third on those indices, and OpenAI's model eighth and fourteenth. TechCrunch also says TypeSafe's models are the most widely queried on OpenRouter.

Nikunj Handa, the product lead for OpenAI's APIs, told TechCrunch: "The big unlock with Decisions API has been the speed, shrinking things from one to three seconds to 100, 120, 180 milliseconds." He added: "But right now, it’s just plain old Luna that gets the job done."

OpenAI first announced the API at its September 29 event; see our DevDay 2026 recap and what the launches mean for developers.

#What is still unclear?

  • Exact cache-write prices for Decisions. The pricing page lists them only for gpt-6-luna in general.
  • The regional processing premium. OpenAI's pricing page applies a 10% uplift to models released on or after March 5, 2026, and no page we read gives Luna's release date.
  • Rate limits for the Decisions endpoint, which the pages we read do not state.
  • Benchmarks. The rankings above are TechCrunch's reporting; we have not checked the indices ourselves.