28creation
Newsletter
News AI

OpenAI Decisions API leaves beta with no charge for output

Classifiers and routers can get typed answers from gpt-6-luna for $0.10 per million input tokens. The saving is the output fee; the bigger change is speed.

By 28creation Desk 3 min read

The short read

OpenAI's Decisions API charges nothing for output tokens and claims answers about 10x faster than the Responses API.

  • The API left beta on October 22. It runs only gpt-6-luna, takes text, images and JSON, and returns probabilities and choices rather than free text.

What you can do with this

If you run a yes-or-no or pick-one classifier on gpt-6-luna today, price one day of its traffic on both endpoints. Decisions saves the output fee, but it cannot read files or audio.

Hands typing on a laptop with code on the screen, a mug beside it.
Photo: Free-Photos (Pixabay), CC0

OpenAI’s Decisions API is now generally available, its API changelog says in an October 22 entry. The guide prices uncached input at $0.10 per million tokens, and output is not charged at all. If your product routes tickets or scores leads with a language model, those calls can now run cheaper and faster than on the Responses API, with limits worth checking first.

The API had been in public beta since October 6. TechCrunch reported the launch the same day.

A classifier with typed answers

The endpoint is POST /v1/decisions, and gpt-6-luna is “the only currently supported model,” the guide says. Instead of a prompt that asks for text, a request carries up to 400 questions, up from 200 in the beta. Each question has one of three types: a predicate returns a probability from 0 to 1, a choice returns one of your options with a probability per option, and a score rates the item on a scale of 2 to 10 levels.

The guide claims answers “about 10x faster than the Responses API.” OpenAI did not train a new model for this. TechCrunch reports the team reconfigured Luna for faster inference, and quotes Nikunj Handa, product lead for OpenAI’s APIs: “it’s just plain old Luna that gets the job done.”

Handa told TechCrunch latency fell from one to three seconds to “100, 120, 180 milliseconds.” Box, Datadog and Ramp were among its early trial users, the report says.

The price cut is the output fee

The input price is no discount. OpenAI’s pricing page lists standard gpt-6-luna rates at the same $0.10 per million input tokens, plus $0.50 per million output tokens and $0.01 for cached input. The guide says Decisions cache reads match those caching rates, and cache writes are charged.

A hypothetical triage job shows the size of it: 10,000 emails of 800 tokens each is 8 million input tokens, or $0.80 on either endpoint. If Responses wrote 40 output tokens per email, that adds 400,000 tokens, or $0.20. Decisions would cost $0.80 against $1.00. Short answers keep the saving modest; speed is the stronger reason to switch.

What it will not do

The trade-offs are in the guide. “Files, audio, and item references aren’t supported,” so PDFs need converting first. When you need an object in your own JSON schema, the guide points you back to Structured Outputs on the Responses API.

Voice works only through GPT-Live, which, per OpenAI’s voice guide, “sends the conversation transcript to Decisions automatically.” Regional processing premiums and long-context multipliers also apply. And on two outside rankings cited by TechCrunch, OpenAI’s decision model placed 8th and 14th, behind rival Jev from TypeSafe at 7th and 3rd.

What to do now

  1. List your classifier calls. A prompt that ends in yes or no maps onto a predicate question; one that picks a label maps onto a choice.
  2. Price a real day. Take one day of input and output tokens from your usage export and apply $0.10 input with no output fee, against $0.10 and $0.50 on Responses.
  3. Batch questions per item. One request can ask up to 400 questions about the same email, so ask them together.
  4. Keep a shared prefix of 1,024 tokens or more. Implicit caching needs it, and question text does not count toward it.
  5. Track it separately. Usage is reported at GET /v1/organization/usage/decisions with an admin key that has api.usage.read.

Sources

Read next

  1. News Work software

    Nextcloud Collabora flaw joins CISA list; 29 and older get no fix

    Attackers use a flaw in Nextcloud's built-in Collabora server to plant web shells. Nextcloud 29 and older have no patched build: upgrade or remove the app.

    28creation Desk 3 min read

  2. News Developer tools

    Brotli flaw spreads past Node.js to Python, Envoy, Istio and Bun

    CVE-2026-111497 sits in Google's Brotli decoder, so any service that unpacks Brotli data needs a patched build, not only Node.js apps.

    28creation Desk 3 min read

  3. News Developer tools

    eMudhra quits TLS after key flaw; old trust stores still at risk

    Browsers already block eMudhra's eight broken CAs. Servers, scripts, containers and macOS command-line tools may still trust them until you update.

    28creation Desk 4 min read