Decisions API is in public beta: get typed yes/no, pick-one, or score answers about 10x faster than Responses

#What changed

OpenAI opened the Decisions API to all developers in public beta on Oct 6. Tibo (@thsottiaux) posted it as Day 2.4 of the Codex 28-day program and said OpenAI will use it in its own apps, calling a realtime general classifier a useful building block. OpenAI Developers says it makes decisions up to 10x faster than calling GPT-6 Luna through the Responses API. OpenAI's Decisions guide describes a request with three parts: a model (only gpt-6-luna for now), an input (a text string, or user messages with text and images), and a list of questions. Each question is one of three types. A predicate returns the probability that a condition is true. A choice returns one of the values you supplied, plus probabilities and a confidence. A score rates the input against ordered levels you define and returns a probability-weighted score. The guide says general availability is expected in the coming weeks.

#Who is affected

Developers who call a chat model today just to classify, route, filter, or triage something, such as routing support tickets, sorting a moderation queue, picking an agent's next action, or checking a product photo for damage. This is API only. Nothing here changes ChatGPT plans or Codex subscription usage.

#What to do now

• Try your questions in the Decisions playground at platform.openai.com/decisions before writing code.
• Move calls that only need a label or a yes/no over to POST /v1/decisions. Keep Structured Outputs or function calling for anything that has to generate fields, text, or tool arguments; the guide draws that line itself.
• Add a fallback choice such as "other" so inputs outside your categories go to a review queue.
• Set thresholds from labeled examples of your own data, based on what a false positive or a false negative costs you, rather than trusting a raw probability.
• Put independent questions in one request, and send dependent ones as separate calls.
• Send images as inline base64 data URLs. Hosted image URLs and file_id inputs aren't accepted by this endpoint.
• Budget for input only: $0.10 per 1M input tokens, with no output or caching charges. Regional processing and long-context multipliers still apply.

#What is not confirmed

• The 10x speed figure comes from OpenAI. We haven't measured latency ourselves.
• The guide doesn't spell out beta rate limits for /v1/decisions.
• Tibo didn't say which ChatGPT or Codex features will use it, or when.
• Several unrelated third-party services also call themselves a "Decisions API". Only OpenAI's developer docs describe OpenAI's endpoint.