OpenAI announced a new Decisions API on Oct 7, 2026. The service reduces complex evaluations to simple yes/no, pick‑one, or rating responses. It works with text, images, or both and is currently in public beta, with general availability promised soon. Only the gpt‑6‑luna model is supported at launch.
How the API works and who can use it
The API returns three kinds of outputs: a probability for yes or no, a choice from a list you define, or a score on a numeric scale. OpenAI says it can be used for things like spotting damage in photos, routing customer inquiries automatically, and classifying documents. It runs about ten times faster than OpenAI’s existing Responses API, which should lower latency for real‑time apps. The service stores no data after processing and meets HIPAA requirements in the US and Europe, making it suitable for health‑related workflows.
Pricing and tier changes
OpenAI priced the Decisions API at $0.10 per million input tokens, and output tokens are free. At the same time, the company trimmed its paid API plans from five to three – Build, Launch, and Grow – with automatic upgrades as usage hits higher credit thresholds. Monthly spend caps are $500 for Build, $5,000 for Launch, and $200,000 for Grow.
Why it matters
Businesses can now add quick decision‑making to apps without building their own models, cutting both latency and cost. The HIPAA‑compliant design opens the service to health‑care use cases that need fast, secure classification. It remains unclear how independent tests will compare its accuracy to existing solutions.
