Overview
- OpenAI declared the Decisions API generally available on October 22, 2026, and made it accessible to eligible developers in select global markets through the Decisions Playground.
- The API runs only on gpt-6-luna and accepts up to 400 independent questions per request, a limit raised from the public beta to support larger batch decisioning.
- OpenAI bills only for input tokens at $0.10 per 1M uncached tokens and $0.01 per 1M cached tokens, and it does not charge for output tokens; prompt caching is highlighted as the main lever to lower costs and latency.
- GA added native GPT‑Live integration, image and native JSON inputs, regional processing options, Zero Data Retention and HIPAA eligibility, and OpenAI and early customers reported faster latencies (vendor demos cited ~150 ms and up to 10× speed gains) that users should validate on their own workloads.
- The product is aimed at hot‑path tasks like routing, classification and micro‑decisions for agents and live voice loops, and developers are advised to compare latency, cost and residency needs against alternatives such as Microsoft’s Decision‑1 before adopting it.