What OpenAI Announced
On October 6, 2026, OpenAI moved the Decisions API from limited preview (first shown at DevDay on September 29) into public beta. It is a dedicated endpoint — POST /v1/decisions — that returns structured, typed answers to questions you define about text or image input, instead of generated prose. The only model behind it in beta is GPT-6 Luna.
Three Answer Types, Zero Prose
Every request bundles evidence with named questions, and each answer comes back in one of three shapes: a probability that a condition is true (0 to 1), one choice from a fixed list you supply, or a score against ordered levels such as severity. That removes the classic loop of asking a chat model for a paragraph and regexing out a true/false.
Speed and Pricing
OpenAI's headline claim is speed: answers come back about 10x faster than asking the same question through the Responses API — its own benchmark, not yet independently verified. Pricing is $0.10 per million input tokens, with no charge for output tokens, cache reads, or cache writes. Regional processing premiums and long-context multipliers can still apply.
- Probability: estimate how likely a condition is true, e.g. whether a support ticket is urgent
- Choice: pick one option from a list you define, e.g. which queue a ticket belongs to
- Score: rate the input against ordered levels, e.g. damage severity in a photo
Image Rules and Compliance
The endpoint accepts both text and images, but images must be inline base64 data URLs — hosted HTTP/HTTPS URLs and uploaded file IDs are refused, which keeps server-side fetch latency at zero. The beta also ships with Zero Data Retention support, HIPAA eligibility, and US plus EU data residency, making it a plausible drop-in for moderation and gating flows.
| Decisions API pricing | Value |
|---|---|
| Input tokens | $0.10 per 1M |
| Output tokens | $0.00 |
| Cache reads | $0.00 |
| Cache writes | $0.00 |
| Regional premiums | still apply |
Who Should Care
If you build anything that routes, triages, classifies, or gates — support ticket routing, content moderation queues, damage detection from photos, next-action selection in agents — the Decisions API replaces a fragile prompt-and-parse pipeline with one typed call. The intake, redaction, thresholds, audit log, and human review queue are still yours to build; this is the verdict layer, not the workflow.
Frequently Asked Questions
What is the OpenAI Decisions API?
It is a public-beta endpoint (POST /v1/decisions, launched October 6, 2026) that returns typed answers — probabilities, choices, or scores — about text or image input, powered only by GPT-6 Luna. OpenAI says it is about 10x faster than the Responses API for this kind of work.
How much does the Decisions API cost?
$0.10 per million input tokens, with no charges for output tokens, cache reads, or cache writes. Regional processing premiums and long-context input multipliers can still apply.
Can I pass an image URL to the Decisions API?
No. Hosted image URLs and file IDs are not supported; images must be sent inline as base64 data URLs, with a limit of 128 images per request.
When will the Decisions API reach general availability?
OpenAI says GA is expected in the coming weeks after the October 6, 2026 public beta. A free playground exists today for testing.
Is the Decisions API compliant for regulated use?
The beta supports Zero Data Retention, HIPAA eligibility, and data residency in the US and Europe (EEA/Switzerland), which suits moderation and gating flows in regulated teams.
