OpenAI Decisions API

Key facts: The OpenAI Decisions API, generally available from October 22, 2026, is built for a different job than a chatbot: it takes a case — a transaction, a support ticket, a piece of content — and returns a structured, typed verdict rather than prose. That makes it a natural fit for fraud checks, moderation, approvals and triage, where software needs an answer it can act on, not a paragraph to parse.

Plans: the API is priced at $0.10 per million input tokens with no charge for output tokens, according to OpenAI’s launch pricing — roughly an order of magnitude cheaper than running the same decision through a full chat model. OpenAI describes it as about ten times faster than its Responses API for these verdict-style calls; treat vendor speed claims as unverified until independent tests land.

Best for: developers automating decisions at scale — risk, trust and safety, and operations teams. Watch out: a verdict API is only as good as the policy you encode in it, and every automated decision needs an appeal path for the humans on the other end.

Pros

• Returns a structured, typed verdict instead of prose, so software can act on it directly.
• Built for fraud checks, moderation, approvals and triage where a clear answer matters.
• Generally available from October 22, 2026, as a dedicated decision-making API.

Watch out for

• It is a developer API, not a chatbot; it needs engineering work to integrate.
• Verdicts are only as good as the case data and rules you feed it.
• High-stakes decisions still need human oversight and appeal paths.

Best for

Developers building fraud checks, moderation, approvals and triage into software. Who should skip it: anyone wanting open-ended chat or creative text should use a conversational model instead.

Pricing

Pricing is usage-based; check the official pricing page for current plans and limits.

FAQs

What does the Decisions API return? A structured, typed verdict rather than an essay.

What is it used for? Fraud checks, moderation, approvals and triage, where software needs an answer it can act on.

Is it a chatbot? No, it is built for a different job than a chatbot: machine-readable decisions.