OpenAI Decisions API: fast, cheap decisions with GPT-6 Luna

What changed?

On October 22, OpenAI made the Decisions API generally available (OpenAI). It does not generate text: you define a question and its allowed answers, pass context as text, images or JSON, and get a structured decision back. It is built on GPT-6 Luna, and OpenAI says it is up to 10x faster than the same model through the Responses API.

New since the beta

• Native GPT-Live integration, so decisions can run during live voice sessions via session.triggers.
• Prompt caching for reused context.
• Inference residency in the US and EU.
• Native JSON inputs and images in question instructions.
• Up to 400 questions per request, twice the previous limit.

Three answer types

• predicate: a 0 to 1 probability that a condition is true, such as whether a product shows visible damage.
• choice: one of your predefined options, plus probabilities and confidence, such as whether an agent should continue, retry or ask for clarification.
• score: a rating against ordered rubric levels, returned as the probability-weighted average of level indices.

Pricing and customer numbers

Input costs $0.10 per million tokens, cached input $0.01, and there are no output-token charges. Zero Data Retention is available to eligible customers. Customers quoted by OpenAI report 40x faster p95 at comparable accuracy (Datadog), under 280 ms p95 with 90% lower cost per call (Sierra) and 4x faster at 30% lower cost (Ramp). These are the companies' own measurements; there is no independent benchmark yet.

Why it matters

Decision models have so far been a startup category (Jev, Liquid d1, Perplexity Decider, Baseten mammoth-1). OpenAI is now in, at a price that undercuts small models. If your agent runs many small routing or classification calls through a full LLM, move one recurring decision here and benchmark it on your own data.